KR20180115448A - Apparatus and method for authenticating time-varying signal in online via kernel regression model - Google Patents

Abstract

According to the present invention, an apparatus and a method for authenticating a time varying signal in online via a kernel regression model. According to an embodiment of the present invention, the apparatus for authenticating online comprises: a data transforming unit for converting memory data by using a time dependent conversion transform based on a difference between a first time and a second time; a query vector providing unit for providing a query vector based on the transformed memory data; a distance calculating unit for calculating a Euclidean distance for the query vector by using the transformed memory data; a weighted value calculating unit for calculating a kernel weighted value by applying a Gaussian kernel function for the calculated Euclidean distance; and a response variable calculating unit for calculating a response variable for the query vector by using the calculated weighted value.

Description

Field of the Invention [0001] The present invention relates to an online verification apparatus and method for verifying a time-varying signal through a kernel regression model,

The present invention relates to a technical idea for on-line verification of a signal that varies with time using a kernel regression model, and an on-line verification of a time-varying signal related to a process of an industrial facility by modifying a conventional kernel regression model based on a time interval And more particularly,

An industrial facility is a device made up of a number of systems and devices to achieve a specific purpose. Generally, one or more instruments are installed to check the operational and safety status and can be measured offline or online.

The efficiency and stability of equipment can be changed according to the external conditions (atmospheric temperature, pressure, humidity, temperature of sea water or precipitation when cooling water is required), the characteristics of the input fuel,

Online monitoring and signal validation techniques are preferred due to the demand for robust and resilient performance to monitor these process and instrument parameters. Online monitoring and signal validation techniques include automated methods to monitor device performance during operation of industrial facilities (eg, power plants).

At least one empirical model is used to implement the technology, one of which is a nonparametric regression model, commonly known as kernel regression. Unlike the general regression model, which requires parameter estimation, the kernel regression performs an algorithm estimation procedure without assuming the assumption of parameters.

The kernel regression algorithm can take into account multiple variables such as the following equations (1) and (2).

[Equation 1]

According to Equation (1), X can represent a prediction variable, m can represent the number of memory vectors, and p can represent the number of prediction variables.

For example, x _ij can represent the ith observation of the jth prediction variable.

&Quot; (2) "

According to Equation (2), y may represent a response variable, and m may represent the number of memory vectors.

For example, the Euclidean distance, the kernel weight (Gaussian kernel), and the expected response variable for the query vector observed based on equations (1) and (2) are given by Equation 4, Equation 5 and Equation 6, respectively Can be calculated on the basis of. First, the query vector may be expressed by the following equation (3).

&Quot; (3) "

In Equation 3, x _q can represent a query vector, p may represent the number of predictors.

&Quot; (4) "

In Equation (4), d _i may represent the i-th Euclidean distance, x _i may represent the i-th predictor observation, and x _q may represent the query vector.

&Quot; (5) "

는 커널 대역폭을 나타낼 수 있다.In Equation (5), K _i may represent an i-th kernel weight, X _i may represent an i-th predicted variable, x _q may represent a query vector, and d _i may be an i-th Euclidean distance Lt; / RTI >

Can represent the kernel bandwidth.

For example, the value of the kernel weight may range between "0" and "1 ". A value of "1" may indicate that two compared vectors are equal, and a value approaching "0" or "0" may indicate a distance or non-similarity between the two vectors.

&Quot; (6) "

는 예상 응답 변수의 가중 평균을 나타낼 수 있고, K는 커널 가중치를 나타낼 수 있고, X_i는 i번째의 예상 변수를 나타낼 수 있으며, x_q는 쿼리 벡터를 나타낼수 있고, y_i는 i번째의 응답 변수를 나타낼 수 있다.In Equation (6)

May represent the weighted average of the expected response variables, K may represent the kernel weights, X _i may represent the i-th predictor variable, x _q may represent the query vector, y _i may represent the query vector, Response variables can be represented.

For example, equation (6) can be used to estimate predictive variables in inferential KR (kernel regression), auto-associative KR (AAKR), or hetero-associative form.

Kernel regression has been used to build many data-driven models and has proven to be efficient in terms of applications. Kernel regression is a process of estimating parameter values by statistically modeling the weighted average of past observations. On the other hand, kernel regression can be represented by the Nadraya-Waston estimator.

In addition, in the past, the facility data and equipment abnormality with definite linearity have been defined, and the relationship has been managed by using various statistical methods. However, when the relationship between the equipment data and the equipment abnormality does not necessarily have a linearity, it is difficult to predict the equipment abnormality because it is difficult to find a relationship with the statistical method when the equipment data and the equipment abnormality have nonlinearity. It was difficult to cope with poetry.

 General facility data does not show uniformity and uniform distribution, so various non-parametric methodologies have to be found, but it is time-consuming and difficult to increase reliability.

That is, when the variable is estimated through the conventional kernel regression, the time for estimating the variable is changed, but the first predicted point and the second predicted point are measured in the same manner, and the Euclidean distance is the distance to the two data vectors It is not possible to distinguish two data vectors by computing the same value.

Therefore, when the repeated search is performed, there is a disadvantage in that the result predicting place of the first predicted point and the second predicted point can take the same value as the average of the expected expected values.

Korean Patent No. 10-0997009, "Dynamic monitoring and timely alarm method for the process margin of industrial equipment" Korean Patent No. 10-0867938, " Prediction Method for Performance Monitoring of Power Plant Instruments Using Dependent Variable Similarity and Kernel Regression Method " Korean Patent Laid-Open No. 10-2015-0129507, "Method and System for Establishment of Equipment Abnormal Prediction Model" Korean Patent No. 10-1096793 entitled " Method for collecting data for a process margin monitoring system of an industrial facility and its storage medium "

The present invention provides an apparatus and method for online verification of time-varying signals through a kernel regression model.

The present invention aims to provide an apparatus and method for improving the accuracy of on-line verification of time-varying signals by modifying the kernel regression model in consideration of time intervals.

SUMMARY OF THE INVENTION The present invention seeks to provide an apparatus and method for improving the accuracy of on-line verification of time-varying signals by modifying the kernel regression model in consideration of the timing information and the influence of previous data points on changes in the current data points.

The present invention seeks to provide an apparatus and method for deriving a timing dependent transformation equation that incorporates timing and information of previous data points into an evaluation of similarity metrics of current data points.

The present invention provides an apparatus and method for deriving a time-dependent conversion formula in consideration of a time interval and converting memory data using an induced time-dependent conversion formula.

SUMMARY OF THE INVENTION The present invention aims to provide an apparatus and method for correcting and verifying an accurate signal in time-varying data by compensating for the limitation on the time-varying signal.

The present invention provides an apparatus and method for improving operational performance of a power plant by providing early warning information to an operator performing accurate signal verification and efficient monitoring in time-varying data.

According to an embodiment of the present invention, an online verification apparatus includes a data conversion unit for converting memory data using a time-dependent conversion formula based on a difference between a first time and a second time, A distance calculator for calculating a Euclidean distance with respect to the query vector using the converted memory data, a Gaussian kernel function applied to the calculated Euclidean distance to calculate a kernel weight And a response variable calculator for calculating a response variable for the query vector using the calculated weight value.

According to an embodiment of the present invention, the on-line verification apparatus multiplies a value obtained by subtracting the first time from the second time by a tangential differential value, adds a first predictive variable at the first time, The second predictive variable of the time-dependent conversion equation may be calculated to derive the time-dependent conversion equation.

According to an embodiment of the present invention, the on-line verification apparatus may calculate a first predicted value of a first data point related to the first time by multiplying a value obtained by multiplying a tangential differential value of a second data point related to the second time with the second time, By adding a value obtained by subtracting a first conversion result converted from a first predicted variable of the first data point from the second data point to convert the memory data.

According to an embodiment of the present invention, an on-line verification apparatus combines timing generation information associated with a first predictor variable of the first data point with a second predictor variable of the second data point using the derived time- The memory data can be converted.

According to an embodiment of the present invention, the on-line verification apparatus may provide the query vector reflecting timing generation information of the first data point.

According to an embodiment of the present invention, the online verification apparatus can calculate the Euclidean distance between the first data point and the second data point using the query vector.

According to an embodiment of the present invention, the online verification apparatus can calculate the kernel weight using the Gaussian kernel function for the calculated Euclidean distance and kernel bandwidth.

According to an embodiment of the present invention, an on-line verification apparatus calculates a weighted average of response variables corresponding to the query vector based on the calculated kernel weight, the Euclidean distance, and a response variable corresponding to the second predictor variable Can be calculated.

According to an embodiment of the present invention, the on-line verification apparatus may further include a process determination unit for monitoring a signal corresponding to the query vector based on the weighted average to determine a process margin of an industrial facility.

According to an embodiment of the present invention, the on-line verification apparatus can finite-difference the at least three data points among the converted memory data to approximate the converted memory data.

According to one embodiment of the present invention, the on-line verification apparatus further includes a path selection unit for providing either a training path or an execution path of the kernel regression model, and a hysteresis data collection unit for collecting hysteresis data related to operation of the power plant during operation of the power plant can do.

According to one embodiment of the present invention, an on-line verification method includes a step of converting, in a data conversion unit, memory data using a time-dependent conversion formula based on a difference between a first time and a second time, Calculating a Euclidean distance for the query vector using the transformed memory data in a distance calculator, and calculating a weighted sum of the Euclidean distances for the query vector, Calculating a kernel weight by applying a Gaussian kernel function to the dian distance, and calculating a response variable for the query vector using the calculated weight in the response variable calculator.

According to an embodiment of the present invention, an on-line verification method includes multiplying a value obtained by subtracting the first time from the second time by a tangential differential value, adding a first predictive variable at the first time, And deriving the time-dependent conversion equation by calculating a second predictive variable of the time-dependent conversion equation.

According to an embodiment of the present invention, an on-line verification method includes comparing a first predicted value of a first data point related to the first time with a value obtained by multiplying a tangential differential value of a second data point related to the second time with the second time, And calculating a second conversion result to be converted from the second data point by adding a value obtained by subtracting a first conversion result transformed from a first predictive variable of the first data point.

According to an embodiment of the present invention, an on-line verification method combines timing generation information associated with a first predictor variable of the first data point with a second predictor variable of the second data point using the derived time- Step < / RTI >

According to an embodiment of the present invention, the online verification apparatus can improve the accuracy of on-line verification of the time-varying signal by modifying the kernel regression model in consideration of the time interval.

Also, according to one embodiment of the present invention, the on-line verification apparatus can improve the accuracy of on-line verification of time-varying signals by modifying the kernel regression model in consideration of timing information and influence of previous data points on a change of a current data point .

In addition, according to an embodiment of the present invention, the on-line verification apparatus can derive a time-dependent conversion formula that integrates timing and information of previous data points with evaluation of a similarity metric of the current data point.

Also, according to an embodiment of the present invention, the online verification apparatus can derive a time-dependent conversion formula in consideration of a time interval, and convert memory data using an induced time-dependent conversion formula.

Also, according to an embodiment of the present invention, the on-line verifying apparatus can correct the limit of the time varying signal and perform accurate signal verification and efficient monitoring in time varying data.

Also, according to one embodiment of the present invention, the online verification apparatus can provide early warning information to an operator performing accurate signal verification and efficient monitoring in time-varying data, thereby improving the operation performance of the power plant.

1 shows a block diagram of an on-line verification apparatus according to an embodiment of the present invention.
Figure 2 shows a graph associated with predicted vs. contrast response variables through a kernel regression model according to the prior art.
Figure 3 shows a graph associated with the derivation of a time dependent conversion equation in accordance with an embodiment of the present invention.
4 shows a block diagram of a kernel regression model according to an embodiment of the present invention.
Figure 5 shows a flow diagram associated with an on-line verification method in accordance with an embodiment of the present invention.
Figure 6 shows a flow diagram associated with an on-line verification method according to another embodiment of the present invention.

Hereinafter, various embodiments of the present document will be described with reference to the accompanying drawings.

It is to be understood that the embodiments and terminologies used herein are not intended to limit the invention to the particular embodiments described, but to include various modifications, equivalents, and / or alternatives of the embodiments.

In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail.

The following terms are defined in consideration of functions in various embodiments and may vary depending on the intention of a user, an operator, or the like. Therefore, the definition should be based on the contents throughout this specification.

In connection with the description of the drawings, like reference numerals may be used for similar components.

The singular expressions may include plural expressions unless the context clearly dictates otherwise.

In this document, the expressions "A or B" or "at least one of A and / or B" and the like may include all possible combinations of the items listed together.

Expressions such as " first, "" second," " first, "or" second, " But is not limited to those components.

When it is mentioned that some (e.g., first) component is "(functionally or communicatively) connected" or "connected" to another (second) component, May be connected directly to the component, or may be connected through another component (e.g., a third component).

As used herein, the term "configured to" is intended to encompass all types of information, including, but not limited to, " , "" Made to "," can do ", or" designed to ".

In some situations, the expression "a device configured to" may mean that the device can "do " with other devices or components.

For example, a processor configured (or configured) to perform the phrases "A, B, and C" may be implemented by executing one or more software programs stored in a memory device or a dedicated processor (e.g., an embedded processor) , And a general purpose processor (e.g., a CPU or an application processor) capable of performing the corresponding operations.

Also, the term 'or' implies an inclusive or 'inclusive' rather than an exclusive or 'exclusive'.

That is, unless expressly stated otherwise or clear from the context, the expression 'x uses a or b' means any of the natural inclusive permutations.

1 shows a block diagram of an on-line verification apparatus according to an embodiment of the present invention. Specifically, FIG. 1 illustrates components of an on-line verification apparatus according to an embodiment of the present invention. Less used '. Partial ','. Quot; and the like denote a unit for processing at least one function or operation, and may be implemented by hardware, software, or a combination of hardware and software.

1, the online verification apparatus 100 includes a data conversion unit 110, a query vector providing unit 120, a distance calculation unit 130, a weight calculation unit 140, and a response variable calculation unit 150 .

According to an embodiment of the present invention, the data conversion unit 110 may convert the memory data using a time-dependent conversion formula based on the difference between the first time and the second time.

That is, the data conversion unit 110 derives a time-dependent conversion formula considering a difference between a first time and a second time corresponding to a time interval between the first data point and the second data point, Memory data can be converted. For example, the memory data may include historical data collected from the plant equipment of the power plant.

According to an example, the data conversion unit 110 multiplies the value obtained by subtracting the first time from the second time by the tangential differential value, and then adds the first predictive variable at the first time to the second predictor at the second time A time dependent conversion equation can be derived by calculating the variables. For example, a tangential derivative may include a derivative.

In addition, according to an embodiment of the present invention, the data conversion unit 110 multiplies the value obtained by multiplying the tangential differential value of the second data point related to the second time by the second time to the first time point of the first data point related to the first time The memory data can be converted by calculating a second conversion result that is converted from the second data point by adding a value obtained by subtracting the first conversion result converted from the first predicted variable of the first data point in the predicted variable.

According to an example, the data conversion unit 110 may convert the timing generation information associated with the first predicted variable of the first data point to the second predicted variable of the second data point using the time- To convert the memory data. For example, the first data point may be located before the second data point with respect to the time axis.

Also, according to an embodiment of the present invention, the data conversion unit 110 may perform finite-difference on at least three or more data points of the converted memory data to approximate the converted memory data .

According to an embodiment of the present invention, the query vector providing unit 120 may provide a query vector based on the converted memory data using a time-dependent conversion formula.

Also, according to an embodiment of the present invention, the query vector providing unit 120 may provide a query vector reflecting the timing generation information of the first data point.

In one example, the query vector providing unit 120 may provide a query vector corresponding to a particular observation value of a particular one of the transformed memory data.

According to an embodiment of the present invention, the distance calculator 130 may calculate the Euclidean distance to the query vector using the transformed result from the memory data.

For example, the distance calculator 130 may calculate the Euclidean distance between the first data point and the second data point using the query vector. That is, the Euclidean distance can be calculated in consideration of the time interval between the first data point and the second data point.

According to an embodiment of the present invention, the weight calculation unit 140 may calculate the kernel weight by applying a Gaussian kernel function to the Euclidean distance calculated by the distance calculation unit 130. [

In addition, for example, the weight calculation unit 140 may calculate the kernel weight using the Gaussian kernel function for the calculated Euclidean distance and the kernel bandwidth.

According to an embodiment of the present invention, the response variable calculator 150 may calculate a response variable for the query vector using the calculated weight. For example, the response variable may include a dependent variable.

For example, the response variable calculator 150 may use a kernel weight, a Euclidean distance between the first data point and the second data point, and a response variable for the second predictor variable corresponding to the second data point, The weighted average of the response variables can be calculated.

According to another embodiment of the present invention, the online verification apparatus 100 may include a process determination unit (not shown), a path selection unit (not shown), and a history data collection unit (not shown).

A process determination unit (not shown) according to another embodiment of the present invention can determine a process margin of an industrial facility by monitoring a signal corresponding to a query vector based on a weighted average.

For example, the process judging unit (not shown) determines that the weighted average is between "0" and "1", and judges that the similarity is higher when the determined value is closer to "1" can do.

A path selector (not shown) according to another embodiment of the present invention may provide either a training path or an execution path of a kernel regression model. That is, a training path that a path selecting unit (not shown) can generate a training model is selected, and a path selecting unit (not shown) may select an execution path for performing monitoring using the generated training model.

A history data collection unit (not shown) according to another embodiment of the present invention may collect historical data related to operation of the power plant during operation of the power plant. That is, the historical data collection unit (not shown) may collect historical data for performing dynamic monitoring of the process margin and timely alarm using the operation data of the industrial facility located at the power plant. Also, the history data may include learning data.

For example, the learning data is derived from the operating conditions of a normal industrial plant without any data deterioration or efficiency deterioration, and is based on external conditions (air temperature, pressure, humidity, sea water or precipitation temperature when cooling water is required) Range, and the like. All variables or some variables that can be measured by on / off-line method can be treated as learning data through the measurement equipment installed in the monitored industrial facility.

There is no requirement for the collection time of training data. However, in order to collect sufficient data under various conditions as described above, it is preferable to collect data for at least one year while the monitoring target facility is installed for the first time under stable operation. If it is practically impossible to collect more than one year of data in a stable operating condition, an alternative method can be used. For example, it can be replaced by collecting data for three months after the plan preventive suspension has been carried out for many years.

Figure 2 shows a graph associated with predicted vs. contrast response variables through a kernel regression model according to the prior art.

Specifically, FIG. 2 illustrates response variables when inserting predictive variables using a kernel regression model according to the prior art.

Referring to FIG. 2, the horizontal axis of the graph may represent time, and the vertical axis may represent a variable. The variable includes a first predicted variable 210, a second predicted variable 212, a first response variable 222 and a second response variable 224, the time including a first time 230 and a second time 232).

Here, the prior art kernel regression model uses a first predictive variable 210 and a second predictive variable 212 to generate a response variable 200 (including a first response variable 222 and a second response variable 224) ) Can be estimated.

However, in the first time 230 and the second time 232, the first predictive variable 210 and the second predictive variable 212 are the same. The same points include a first data point 240 and a second data point 242.

In this case, as the predicted vector at the first data point 240 is equal to the predicted vector at the second data point 242, the first predicted variable at the first data point 240 and at the second data point 242, The first response variable 222 and the second response variable 224, which correspond to the second predicted variable 212 and the second predicted variable 214, respectively, may not be distinguished.

Also, it is possible that, as a result of repeated investigations, the resultant values for the first response variable 222 and the second response variable 224 are equal to the expected value and the average value, a signal defect may not be identified.

Figure 3 shows a graph associated with the derivation of a time dependent conversion equation in accordance with an embodiment of the present invention.

Specifically, FIG. 3 illustrates a graph associated with a procedure for deriving a time dependent conversion equation using Taylor series expansion according to an embodiment of the present invention.

Referring to FIG. 3, the abscissa of the graph may represent time, and the ordinate may represent a variable. The variable includes a first predictive variable 310, a second predictive variable 312 and a difference value 314 between the first predictive variable and the second predictive variable and the time includes a first time 320 and a second time 332, and includes a first tangent line 330 and a second tangent line 332.

The function at the second predictor variable 312 can be approximated using Equation (7).

&Quot; (7) "

According to Equation 7, x (t ₁₎ may be in the second prediction parameters in a second time can indicate a first predictor in the first _{time, x (t 2), t} 1 is the first time It can represent, and, t ₂ can represent a second time, dx / dt may indicate the tangential derivative.

Here, the time interval may be the difference value of the first time from the second time, and dx / dt is the differential of the tangent line. If the time interval approaches 0, the differential value of the division line through the first and second predicted variables .

Also, the on-line verification apparatus can derive a time-dependent conversion formula using Equation (7). That is, the on-line verification apparatus multiplies the value obtained by subtracting the first time from the second time by the tangential differential value, adds the first predictive variable at the first time, calculates the second predictive variable at the second time, A conversion equation can be derived.

Also, a time-dependent conversion equation for the multivariable function can be calculated as shown in Equation (8) below in accordance with Equation (7).

&Quot; (8) "

는 현재 데이터 포인트의 변환 결과를 나타낼 수 있고,

는

의 이전 변환 결과에 해당하는

와 이전 데이터 포인트에 해당하는

의 차이를 나타낼 수 있으며, t_i는 j번째 변수의 i번째 예측자의 데이터 포인트에 해당하는 x_i,j의 시간 위치을 나타낼 수 있고, dx/dt_i,j는 데이터 포인트 x_i,j에서 접선 미분을 나타낼 수 있다.According to equation (8)

Lt; / RTI > may represent the conversion result of the current data point,

The

Corresponding to the previous conversion result of

And corresponding to the previous data point

, Where t _i may represent the time position of x _{i, j} corresponding to the data point of the i th predictor of the j th variable, and dx / dt _{i, j} may represent the time derivative of the data point x _i, Lt; / RTI >

Further, the online verification apparatus can convert the memory data using Equation (8). That is, the on-line verification apparatus multiplies the value obtained by multiplying the tangential differential value of the second data point related to the second time by the second time with a first prediction of the first data point in the first predictor variable of the first data point related to the first time The memory data can be converted by calculating a second conversion result that is converted from the second data point by adding a value obtained by subtracting the first conversion result converted from the variable.

In addition, the predictive variable can be transformed as shown in Equation (9) based on Equation (8).

&Quot; (9) "

는 변환 결과에 대한 예측 변수를 나타낼 수 있고, m은 변환 결과와 관련된 메모리 벡터의 개수를 나타낼 수 있으며, p는 변환 결과에 대한 예측 변수의 개수를 나타낼 수 있다.According to equation (9)

M can represent the number of memory vectors related to the conversion result, and p can represent the number of prediction variables for the conversion result.

In addition, a query vector may be provided as shown in Equation (10) based on Equation (8).

&Quot; (10) "

는 변환 결과에 기초한 쿼리 벡터를 나타낼 수 있고, p는 상기 변환 결과에 기초한 쿼리 벡터에 대한 예측 변수의 개수를 나타낼 수 있다.According to equation (10)

Can represent a query vector based on the transformed result, and p can represent the number of predicted variables for the query vector based on the transformed result.

Equation (11) can also be used to calculate the Euclidean distance between the first data point and the second data point for the query vector provided based on Equation (10).

&Quot; (11) "

는 i번째 예측 변수로부터의 변환 결과를 나타낼 수 있으며,

는 변환 결과에 기초한 쿼리 벡터를 나타낼 수 있고, p는 상기 변환 결과에 기초한 쿼리 벡터에 대한 예측 변수의 개수를 나타낼 수 있다.According to Equation (11), d _i may represent the Euclidean distance,

Can represent the conversion result from the i-th prediction variable,

Can represent a query vector based on the transformed result, and p can represent the number of predicted variables for the query vector based on the transformed result.

Further, the kernel weight can be calculated using Equation (12) with respect to the Euclidean distance calculated based on Equation (11). Equation 12 is as follows.

&Quot; (12) "

는 커널 대역폭을 나타낼 수 있다.According to Equation (12), Ki may represent the kernel weight associated with the transformed result from the ith predictor, di may represent the Euclidean distance associated with the transformed result from the ith predictor,

Can represent the kernel bandwidth.

Further, the response variable can be calculated based on Equation (13) using the Euclidian distance calculated based on Equation (11) and the kernel weight calculated based on Equation (12).

&Quot; (13) "

는 쿼리 벡터에 해당하는 x_q에 대한 예상 응답 변수의 가중 평균을 나타낼 수 있고, Ki는 i번째의 커널 가중치를 나타낼 수 있으며, di는 i번째의 유클리디안 거리를 나타낼 수 있고, y_i는 i번째의 응답 변수를 나타낼 수 있다.According to Equation (13)

May represent the weighted average of the expected response variable for x _q corresponding to the query vector, Ki may represent the i-th kernel weight, di may represent the i-th Euclidean distance, y _i i-th response variable.

4 shows a block diagram of a kernel regression model according to an embodiment of the present invention.

Specifically, FIG. 4 illustrates a response variable 420 corresponding to a predictor variable 410 passing through a kernel regression model according to an embodiment of the present invention.

When the autocorrelation regression model is constructed with the learning data obtained in a normal state of the facility, a deviation occurs between the input signal and the output signal when the data obtained in the normal state of the facility is input, (parametric) or non-parametric (non-parametric).

For example, an excellent regression analysis model can reduce both goodness-of-fit and smoothness at the same time.

In one example, the kernel regression model 400 may obtain the response variable 420 using the first variable, the second variable, and the m-th variable that the predictive variable 410 includes.

In addition, when the kernel regression model 400 inputs an input signal of a regression analysis model, the regression analysis model performs an autocorrelation analysis to generate an output signal, and the residual is calculated for all variables belonging to the group Respectively, and finally provide the user with the process margin of the facility.

In addition, the process margin can be determined by combining confidence intervals with actual measured values.

Figure 5 shows a flow diagram associated with an on-line verification method in accordance with an embodiment of the present invention.

Specifically, FIG. 5 illustrates a procedure for calculating response variables after an on-line verification method according to an embodiment of the present invention converts memory data using a time-dependent conversion formula considering time intervals between predicted variables.

Referring to FIG. 5, in step 501, the online verification method may convert memory data using a time-dependent conversion formula. That is, the on-line verification method can convert the memory data using the time-dependent conversion formula based on the difference between the first time corresponding to the first prediction variable and the second time corresponding to the second prediction variable.

In step 502, the on-line verification method may provide a query vector. That is, the on-line verification method may provide a query vector for the transformed vectors from the memory data in step 501.

In step 503, the on-line verification method may calculate the Euclidean distance. That is, the on-line verification method can calculate the Euclidean distance to the query vector using the converted vectors.

In step 504, the on-line verification method may calculate the kernel weights by applying a Gaussian kernel function. That is, the on-line verification method can calculate the kernel weight using the Gaussian kernel function for the calculated Euclidean distance and kernel bandwidth.

In step 505, the on-line verification method calculates a response variable for the query vector. That is, the online verification method calculates the response variable for the query vector using the calculated weight. That is, a weighted average of the response variable corresponding to the query vector may be calculated based on the calculated kernel weight, the Euclidean distance, and the response variable corresponding to the second predictor variable.

Figure 6 shows a flow diagram associated with an on-line verification method according to another embodiment of the present invention.

Specifically, FIG. 6 illustrates an example of a procedure for displaying results by signal estimation and error quantization of a query vector through an updated kernel regression model in an on-line verification method of the present invention by selectively operating a training mode and an execution mode do.

Referring to FIG. 6, the on-line verification method determines whether or not the training mode is selected in step 600. Here, if the on-line verification method is the training mode, the process proceeds to step 601, and if the on-line verification method is not the training mode, the process proceeds to step 611. [

In step 601, the online verification method may collect historical data. That is, the on-line verification method can collect historical data related to operation of the power plant during operation of the power plant. Here, the historical data may be related to the operation of the facility of the power plant.

In step 602, the online verification method may derive a time dependent conversion equation. That is, the on-line verification method can derive a time-dependent conversion formula considering the time interval between the first data point corresponding to the first time of the first prediction variable and the second data point corresponding to the second time of the second prediction variable .

In step 603, the on-line verification method may perform approximation using finite difference. That is, the on-line verification method can be approximated by using a finite difference to differentiate a specific data point using a time-dependent conversion formula.

In step 604, the on-line verification method may approximate the memory data using finite difference. That is, the on-line verification method can convert the memory data through approximation to the approximated memory data using the finite difference.

In step 605, the on-line verification method may update the kernel regression model using the converted memory data. That is, the on-line verification method can update the kernel regression model using the result transformed from the memory data.

In step 606, the on-line verification method can verify the status of the training model. In other words, the online verification method judges whether the training model can be applied or not. Here, the online verification method proceeds to step 621 when the training model acquires a score higher than a reference value for judging an appropriate training model, and returns to step 605 to re-update the kernel regression model. Further, in the training mode, steps 621, 622, and 623 may be performed.

In step 611, the online verification method may collect the query vector. That is, the online verification method can collect the query vector from the conversion result from the memory data in the execution mode.

In step 612, the on-line verification method determines whether there are two or more query vectors collected. That is, the on-line verification method determines whether at least three query vectors collected in step 611 are present. The online verification method may return to step 611 if there are three or less query vectors collected.

In step 613, the online verification method may store previous query vectors. That is, the on-line verification method may store the previous query vectors in step 613 while determining according to the number of query vectors collected in step 612. That is, the online verification method does not evaluate the two previous query vectors that are collected but evaluates from the third input query vector.

In step 614, the on-line verification method may perform approximation using finite difference. Also, the on-line verification method can perform the approximation using the following expression (14). Equation (14) is as follows.

&Quot; (14) "

According to Equation 14, dx / dt _{i, j} may indicate the tangential differential approximation in the data point x _{i, j,} x _{i, j} may indicate the i-th observation value of the j-th predictor variable, x _{i -1, j} can represent the (i-1) -th observed value of the j-th prediction variable, x _{i-2, j} can represent the i- The time interval between two hours can be indicated.

In step 615, the on-line verification method may convert the data using the approximated tangent in step 614.

In step 621, the online verification method may update the kernel regression model. That is, the online verification method is determined by updating the kernel regression model.

In step 622, the on-line verification method may quantify the signal estimate and error. That is, the on-line verification method estimates a signal state by estimating a signal associated with a corresponding response variable from a predictive variable, and judges the similarity and non-similarity of signals.

In step 623, the on-line verification method results in a signal associated with the corresponding response variable from the predictor variable.

In step 624, the online verification method inquires the user whether the program is to be terminated. If the termination of the program is selected, the procedure is terminated. If the termination of the program is not selected, the procedure returns to step 611, Lt; / RTI >

In the above-described specific embodiments, elements included in the invention have been expressed singular or plural in accordance with the specific embodiments shown.

It should be understood, however, that the singular or plural representations are selected appropriately for the sake of convenience of description and that the above-described embodiments are not limited to the singular or plural constituent elements, , And may be composed of a plurality of elements even if they are represented by a single number.

While the invention has been shown and described with reference to certain exemplary embodiments thereof, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined by the appended claims.

Therefore, the scope of the present invention should not be limited by the illustrated embodiments, but should be determined by the scope of the appended claims, as well as the appended claims.

100: on-line verification device 110:
120: query vector provider 130: distance calculator
140: Weight calculation unit 150: Response variable calculation unit

Claims (15)

A data converter for converting the memory data using a time-dependent conversion equation based on a difference between the first time and the second time;
A query vector provision unit for providing a query vector based on the converted memory data;
A distance calculation unit for calculating an Euclidean distance to the query vector using the converted memory data;
A weight calculation unit for calculating a kernel weight by applying a Gaussian kernel function to the calculated Euclidean distance; And
And a response variable calculation unit for calculating a response variable for the query vector using the calculated weight value
An on-line verification device that verifies time-varying signals.
The method according to claim 1,
Wherein the data conversion unit comprises:
A second predictive variable at the second time is calculated by adding a first predictive variable at the first time to a value obtained by subtracting the first time from the second time by a tangential differential value, To induce
An on-line verification device that verifies time-varying signals.
3. The method of claim 2,
Wherein the data conversion unit comprises:
Calculating a first predicted variable of the first data point from a first predicted variable of the first data point at a value obtained by multiplying a tangential derivative of the second data point associated with the second time by the second time Calculating a second conversion result to be converted from the second data point by adding a value obtained by subtracting the first conversion result to be converted to convert the memory data
An on-line verification device that verifies time-varying signals.
The method of claim 3,
The data conversion unit
Using the derived time dependent conversion equation to combine timing generation information associated with a first predictor variable of the first data point with a second predictor variable of the second data point to transform the memory data
An on-line verification device that verifies time-varying signals.
5. The method of claim 4,
The query vector providing unit,
The query vector reflecting the timing generation information of the first data point is provided
An on-line verification device that verifies time-varying signals.
6. The method of claim 5,
The distance calculator calculates,
Calculating the Euclidean distance between the first data point and the second data point using the query vector
An on-line verification device that verifies time-varying signals.
The method according to claim 6,
The weight calculation unit may calculate,
The kernel weights are calculated using the Gaussian kernel function for the calculated Euclidean distance and kernel bandwidth
An on-line verification device that verifies time-varying signals.
8. The method of claim 7,
The response variable calculation unit may calculate,
Calculating a weighted average of response variables corresponding to the query vector based on the calculated kernel weight, the Euclidean distance, and the response variable corresponding to the second predictor variable
An on-line verification device that verifies time-varying signals.
9. The method of claim 8,
And a process determining unit for determining a process margin of the industrial equipment by monitoring a signal corresponding to the query vector based on the weighted average
An on-line verification device that verifies time-varying signals.
The method according to claim 1,
Wherein the data conversion unit comprises:
And performing finite-difference on at least three data points of the converted memory data to approximate the converted memory data
An on-line verification device that verifies time-varying signals.
The method according to claim 1,
A path selector for providing either a training path or an execution path of the kernel regression model; And
And a history data collecting unit for collecting historical data related to operation of the power plant during operation of the power plant
An on-line verification device that verifies time-varying signals.
Converting the memory data using a time-dependent conversion equation based on a difference between a first time and a second time;
In the query vector provisioning, providing a query vector based on the transformed memory data;
Calculating a Euclidean distance for the query vector using the transformed memory data;
Calculating a kernel weight by applying a Gaussian kernel function to the calculated Euclidean distance in a weight calculation unit; And
Calculating a response variable for the query vector using the calculated weight in a response variable calculation unit;
An on-line verification method that verifies time-varying signals.
13. The method of claim 12,
Wherein the step of converting the memory data comprises:
A second predictive variable at the second time is calculated by adding a first predictive variable at the first time to a value obtained by subtracting the first time from the second time by a tangential differential value, Lt; RTI ID = 0.0 >
An on-line verification method that verifies time-varying signals.
14. The method of claim 13,
Wherein the step of converting the memory data comprises:
Calculating a first predicted variable of the first data point from a first predicted variable of the first data point at a value obtained by multiplying a tangential derivative of the second data point associated with the second time by the second time And calculating a second conversion result to be converted from the second data point by adding a value obtained by subtracting the first conversion result to be converted
An on-line verification method that verifies time-varying signals.
15. The method of claim 14,
Wherein the step of calculating a second transformation result, which is transformed from the second data point,
And combining timing generation information associated with a first predictor variable of the first data point with a second predictor variable of the second data point using the derived time dependent transformation equation
An on-line verification method that verifies time-varying signals.

Images