KR100867938B1 - Prediction method for watching performance of power plant measuring instrument by dependent variable similarity and kernel feedback - Google Patents

Abstract

A prediction method for a performance measurement of a power plant is provided to improve the accuracy of the prediction value by setting a standard of similarities between measurements and applying the standard to the weighting calculation. A similarity between the measurements is calculated and displayed as a matrix form(ST1). A sum of euclidean with training data about first test data is calculated(ST2). The similarity expressing a linear reliability between measurement signals is calculated and regularized using training data after above process(ST3). The euclidean distances about the first test data are individually obtained(ST4). Weighting value for the test data is calculated using kernel function(ST6). A prediction value for the first test data is obtained by multiplying training data by weighting value(ST7).

Description

Prediction method for watching performance of power plant measuring instrument by dependent variable similarity and kernel feedback}

The present invention relates to a technique for monitoring the performance of the instrument used in the safety monitoring channel of a nuclear power plant at all times while the power plant is operating. In particular, the quantification of the similarity between the instruments and the size of the value is added to the kernel weight calculation. For monitoring the performance of power station instrumentation using dependent variable similarity and kernel regression method, the instrument signal with high similarity is weighted relatively large compared to the instrument signal with low similarity, which is suitable for improving the accuracy of prediction calculation compared to the kernel regression method. It relates to a prediction method.

In general, all power generation facilities are installed in a number of measuring instruments for the purpose of improving operability and securing safety, and acquire signals in real time and use them in power plant monitoring and protection systems. In particular, in order to guarantee the accuracy and reliability of measuring signals, the safety channels of nuclear power plants adopt the concept of multiple instruments and check and calibrate every nuclear cycle (about 18 months) in the operating technical manual. Nuclear power plants around the world are developing technology to extend inspection and calibration cycles by developing condition-based monitoring (CBM) methodologies for instrument calibration that is performed unnecessarily.

Related prior arts include "Non-parametric modeling apparatus and method for classification, especially of activity state" (US Patent, 20070010720), "System for monitoring an industrial or biological process" (US Patent, 5774379), "Kernel-based system" and method for estimation-based equipment condition monitoring "(US Patent, 20050261837)," Diagnostic systems and methods for predictive condition monitoring "(US Patent, 20060036403)," Method and device for signal analysis, process identification and monitoring of a technical process (US Pat. No. 5748508), "Multi-kernel neural network concurrent learning, monitoring, and forecasting system" (US Pat. No. 6216119), "Kernel regression for image processing and reconstruction" (US Pat. No. 20070047838), "Analyzer for modeling and optimizing maintenance operations "(US Pat. No. 6246972).

1 is a block diagram of a performance monitoring system of a general power plant instrument.

As shown in FIG. 1, when the measurement signal is input to the predictive model, the predicted value of the model with respect to the measured value is output. The difference between the measured value and the predicted value is entered into the judgment logic and continuously monitored to detect drift and failure of the instrument.

Since 1992, EPRI has organized the On-Line Monitoring Working Group to develop an Instrument Calibration and Monitoring Program (ICMP).

를 추정한다.The ICMP methodology calculates weights by measuring correlations with other instrument signals in a channel, such as a given instrument signal, in a redundant sensor channel, usually consisting of three or more instruments of the same type at three or more identical locations. True values of multiple instrument signals using

Estimate

Where mi is the signal from instrument I and Ci represents the correlation of the i th signal.

The ICMP calculates an absolute difference obtained by comparing the true value calculated using Equation 1 with the measured value of the multi-measurement channel and determines that there is an error when the threshold value is exceeded. ICMP is based on Catawba and V.C. It is installed and operated successfully in summer nuclear power plant, but it cannot be used in a single channel except for the safety measurement channel, and it can be recognized in advance if any instrument drift occurs in the same direction and with the same performance decay rate. none.

The Argonne National Laboratory developed the Multivariate State Estimation Technique (MSET), obtained US patents, and was commercialized by SmartSignal Corporation and Expert Microsystems for commercial use. Expert Microsystems, Inc. Palo Verde, Limerick 1/2, TMI, V.C. In the Summer, Sequoyah 1 and Salem 1 units, products using MSET are installed on-site to monitor the measuring channel online. Since then, SmartSignal Corporation has not been able to use a patent on MSET, so he developed instrument performance monitoring technology based on kernel regression.

Linear regression is most commonly used to calculate instrument predictive values. This method selects other instrument signals that have a high linear correlation with the instrument signal to be predicted, and calculates a regression coefficient so that the sum of the error squares of the predicted value and the measured value is minimized.

∑E ² = ∑ (Y-Y ') ²

Linear regression analysis can predict independent variables for unknown dependent variables when the regression coefficients are determined by known and independent variables. In the conventional linear regression method, when the dependent variables are linearly related to each other, a problem of multiple collinearity occurs, and a large error occurs in the independent variable for the small noise included in the dependent variable.

Kernel regression does not use parameters such as regression coefficients or weights to optimize the correlation between input and output, like conventional linear regression or neural networks. It is a non-parametric regression method that derives the kernel weight from the Euclidean distance of the training data set in the memory vector for the set and applies it to the memory vector. Nonparametric regression methods, such as kernel regression, have robust advantages over models and signal noise where input / output relationships are nonlinear. The following is the calculation procedure of the existing kernel regression method.

Step 1: Display training data in the form of a matrix.

Where X is the training data matrix stored in the memory vector, n is the number of training data and p is the instrument number

Step 2: Sum the Euclidean distance of the training data for the first instrument signal set.

.

.

.

는 훈련데이터이고,

는 테스트데이터(or Query data), trn은 훈련데이터의 번호,

는 계측기의 번호이다.here

Is training data,

Is the test data, or trn is the number of training data,

Is the number of the instrument.

Step 3: Using the kernel function, find the weights for each training dataset and a given test dataset.

.

.

.

Here we use a Gaussian kernel as the weighting function and it is defined as

Step 4: The predicted value of the test data is obtained by multiplying each training data by the weight and dividing the sum of the weights.

Step 5: Repeat Steps 2 through 4 to obtain predictions for the entire test data.

The previous method of AAKR has a strong advantage over nonlinear models and signal noise, but it does not consider the importance of the variables of individual instruments, so the accuracy is higher than that of the linear regression method for models with a large linear correlation between instrument signals. Falls.

Accordingly, the present invention has been proposed to solve the conventional problems as described above, and an object of the present invention is to quantify the similarity between the instruments, and to further apply the magnitude of the value to the weight calculation of the kernel. The present invention provides a prediction method for power plant instrument performance monitoring using dependent variable similarity and kernel regression that can improve the accuracy of predictive value calculation compared to the conventional kernel regression method by weighting the signal relatively large compared to the instrument signal having a low similarity.

The purpose of monitoring the performance of the measuring instrument used in the nuclear power plant safety monitoring channel of the present invention online while the power plant is operating is to (1) monitor the malfunction of the measuring instrument in real time to improve the reliability of the measuring instrument, and (2) Increase instrument calibration intervals from 18 months up to eight years in current fuel replacement cycles to reduce calibration costs and radiation exposure of calibration workers in the radiation zone, and (3) reduce plant outages due to miscalibration by reducing unnecessary calibrations. And (4) shorten the duration of plant outages to increase plant utilization. The present invention quantifies the similarity between instruments, and further applies the magnitude of the value to the kernel weight calculation to calculate the predicted value compared to the conventional kernel regression method by weighting the instrument signal with high similarity relatively large compared to the instrument signal with small similarity. The purpose is to improve the accuracy of the.

2 is a flowchart illustrating a method for predicting performance monitoring of a power plant instrument using dependent variable similarity and kernel regression according to an embodiment of the present invention.

As shown therein, a first step ST1 of displaying training data in the form of a matrix; A second step (ST2) of obtaining a sum of Euclidean distances with the training data for the first test data after the first step; A third step (ST3) of calculating and normalizing similarity representing a linear relationship between the signals of each instrument using the training data after the second step; A fourth step (ST4) of separately obtaining a Euclidean distance from the training data for the first test data after the third step; A fifth step (ST5) of weighting a similarity matrix to a matrix of Euclidean distances in order to take into account the importance between instrument signals after the fourth step; A sixth step (ST6) of obtaining weights for the given test data for each training data using the kernel function after the fifth step; A seventh step (ST7) of obtaining the predicted value of the first test data given after the sixth step by multiplying the training data by respective weights; And an eighth step (ST8) of repeating the process of the third step to the seventh step until the prediction value for the entire test data is obtained after the seventh step.

In the present invention, a method for predicting the performance of power plant instrumentation using the dependent variable similarity and the kernel regression method quantifies the similarity between the instruments, and applies the magnitude of the value to the kernel weight calculation to add the similarity to the instrument signal having high similarity. It is weighted relatively large compared to the small instrument signal, and thus, it is possible to improve the accuracy of prediction calculations compared to the conventional kernel regression method.

Referring to the accompanying drawings, a preferred embodiment of the method for predicting the performance monitoring of power plant instrument performance using the dependent variable similarity and kernel regression according to the present invention configured as described above is as follows. In the following description of the present invention, detailed descriptions of well-known functions or configurations will be omitted if it is determined that the detailed description of the present invention may unnecessarily obscure the subject matter of the present invention. In addition, terms to be described below are terms defined in consideration of functions in the present invention, which may vary according to intention or precedent of a user or an operator, and thus, the meaning of each term should be interpreted based on the contents throughout the present specification. will be.

First, the present invention stores the selected power plant measurement data in a memory vector, calculates the weight of the kernel from the Euclidean distance of the training data set in the memory vector for the measurement signal set, and applies the same to the memory vector to obtain the predicted value of the instrument. It is about improving the accuracy of kernel regression. To this end, the present invention quantifies the similarity between instruments, and further applies the magnitude of the value to the kernel weight calculation to weight the instrument signal with high similarity to the instrument signal with small similarity, compared to the conventional kernel regression method. Improve the accuracy of predictive value calculation.

The conventional kernel regression method calculates weights using only the Euclidean distances of the training data and the tester data, and calculates the weights of the training data to predict the test data. However, the present invention improves the accuracy of the prediction calculation by including the similarity between the instrument signals as well as the Euclidean distance in the weight calculation.

There are several ways to calculate the correlation or similarity between two instrument signals, and Pearson and Spearman methods are representative.

The first method is the Pearson method, which uses the communism between two signals.

Here is the Spearman method.

Where rank (x _i ) is the order of the i-th instrument signal in order of magnitude.

3 is a table summarized using the Pearson method and the Spearman method for the similarity analysis performance test used in the present invention.

In addition, Figure 4 is a linear data for the similarity analysis performance test used in the present invention, Figure 5 is a reverse linear data for the similarity analysis performance test used in the present invention, Figure 6 is a similarity analysis performance used in the present invention Nonlinear data for testing.

Thus, three types of data such as FIGS. 4, 5, and 6 were used to test the similarity analysis performance of the two methods.

The Pearson method and the Spearman method can measure the exact correlation when the characteristics of the two signals are linear. However, as shown in FIG. 6, when the signals of two instruments are nonlinearly related, the Spearman method may calculate a more accurate correlation than the Pearson method. Therefore, the present invention uses the Spearman method to calculate the mutual similarity of the measurement periods.

1st stage training data in matrix form

Where X is the training data matrix, n is the training data number, and p is the instrument number

-Step 2 Find the sum of Euclidean distance from the training data for the first test data.

.

.

.

는 훈련데이터이고,

는 테스트데이터, trn은 훈련데이터의 개수,

는 계측기의 번호이다.here

Is training data,

Is the test data, trn is the number of training data,

Is the number of the instrument.

Calculate and normalize the similarity representing the linear relationship between each instrument signal using the three-stage training data. s _{i, j} represents the similarity between the meter i and the meter j. The closer the absolute value of this value is to 1, the greater the similarity between the two measurement signals. The closer to 0, the smaller the similarity between the two measurement signals.

Thus, the similarity matrix for the entire instrument signal is

.

.

Step 4: Find the Euclidean distance separately from the training data for the first test data.

.

.

.

The similarity matrix is weighted to the matrix of Euclidean distance to consider the importance between the five-stage instrument signals.

About the first training data

.

.

About the trnth Train data set

.

.

Therefore, the newly obtained Euclidean distance matrix is

Calculate the weight of the given test data for each training data using the 6-stage kernel function.

.

.

Therefore, the weighting function is

Step 7 The prediction of the first test data given is obtained by multiplying the training data by the respective weights.

From here

.

.

-Step 8 Repeat steps 3 to 7 to obtain the predictions for the entire test data.

In order to confirm the effectiveness of the present invention and the superiority of the new methodology, the output of the actual nuclear power plant was increased from 75% to 100%, and compared with the existing methodology using the measurement signal data measured in the secondary system.

7 is a table comparing the accuracy of the conventional kernel regression method and the predicted instrument value of the present invention.

As shown in FIG. 7, the data used in the analysis are measured values for a total of 11 sensors as follows.

-1-2 steam flow rates (klb / hr)

-3 ~ 6 narrow steam generator level (%)

-7 psi gauge in steam generator

-8 wide steam generator water level (%)

-9 ~ 10 main water pipe flow rate (klb / hr)

-11 turbine outputs

FIG. 8 is a nuclear power plant steam generator steam flow rate data for the accuracy test of the present invention in FIG. 7, FIG. 9 is a nuclear power plant steam generator narrow water level data for the accuracy test of the present invention in FIG. 7, and FIG. Nuclear power plant steam generator pressure data for the accuracy test of the present invention, Figure 11 is a nuclear power plant steam generator wide-range water level data for the accuracy test of the present invention in Figure 7, Figure 12 is for the accuracy test of the invention in Figure 7 Nuclear power plant steam generator main water pipe flow rate data, Figure 13 is a nuclear power plant steam generator turbine flow rate data for the accuracy test of the present invention in FIG.

Accuracy is the most basic measure of applying predictive models to driving surveillance. Most of the accuracy is expressed as the mean square error between the model predictions and the actual measurements. The following formula shows the accuracy of a single instrument.

The result of substituting the accuracy calculated by (20) is shown in the table of FIG.

Therefore, the present invention quantifies the similarity between the instruments, and further applies the magnitude of the value to the kernel weight calculation, thereby weighting the instrument signal with high similarity relatively large compared to the instrument signal with small similarity and predicting the value compared with the conventional kernel regression method. This will improve the accuracy of the calculation.

Although the above has been described as being limited to the preferred embodiment of the present invention, the present invention is not limited thereto and various changes, modifications, and equivalents may be used. Therefore, the present invention can be applied by appropriately modifying the above embodiments, it will be obvious that such application also belongs to the scope of the present invention based on the technical idea described in the claims below.

1 is a block diagram of a performance monitoring system of a general power plant instrument.

2 is a flowchart illustrating a method for predicting performance monitoring of a power plant instrument using dependent variable similarity and kernel regression according to an embodiment of the present invention.

3 is a table summarized using the Pearson method and the Spearman method for the similarity analysis performance test used in the present invention.

4 is a linear data for a similarity analysis performance test used in the present invention.

5 is inverse data for the similarity analysis performance test used in the present invention.

6 is nonlinear data for a similarity analysis performance test used in the present invention.

7 is a table comparing the accuracy of the conventional kernel regression method and the predicted instrument value of the present invention.

8 is a nuclear power plant steam generator steam flow rate data for the accuracy test of the present invention in FIG.

9 is a nuclear power plant steam generator narrow level data for the accuracy test of the present invention in FIG.

10 is a nuclear power plant steam generator pressure data for the accuracy test of the present invention in FIG.

FIG. 11 is a nuclear power plant steam generator wide-range water level data for the accuracy test of the present invention in FIG.

FIG. 12 is a nuclear power plant steam generator main water pipe flow rate data for the accuracy test of the present invention in FIG.

13 is a nuclear power plant steam generator turbine flow rate data for the accuracy test of the present invention in FIG.

Claims (12)

Calculating a similarity between the instruments and displaying the training data in the form of a matrix; A second step of obtaining a sum of Euclidean distances with training data for the first test data after the first step; A third step of calculating and normalizing a similarity indicating a linear relationship between each instrument signal using the training data after the second step; A fourth step of separately obtaining a Euclidean distance from the training data for the first test data after the third step; A fifth step of weighting a similarity matrix to a matrix of Euclidean distances in order to take into account the importance between instrument signals after the fourth step; A sixth step of obtaining weights for the given test data for each training data using the kernel function after the fifth step; A seventh step of obtaining the predicted value of the first test data given after the sixth step by multiplying each of the weights by the training data; An eighth step of repeating the processes of the third to seventh steps until a prediction value for the entire test data is obtained after the seventh step; Predictive method for monitoring the performance of power plant instrument using dependent variable similarity and kernel regression, characterized in that performed. The method according to claim 1, The first step is, When calculating the similarity between instruments,

Where rank (x _i ) is the i-th instrument signal value sorted in order of magnitude. The method according to claim 1, The first step is, When displaying training data in the form of a matrix,

Where X is the training data matrix, n is the training data number, and p is the number of the measuring instrument. The method according to claim 1, The second step, When we sum the Euclidean distance with the training data for the first test data,

. . .

As

Is training data,

Is the test data, trn is the number of training data,

Is a number of the measuring instrument, characterized in that the dependent variable similarity and the performance monitoring prediction method for power station instrument using the kernel regression method. The method according to claim 1, The third step, When using the training data to calculate the similarity representing the linear relationship between each instrument signal,

S _{i, j} is a method for predicting performance monitoring of power plant instrument using dependent variable similarity and kernel regression, characterized in that it shows the similarity between instrument i and instrument j. The method according to claim 1, The third step, Using the training data to calculate the similarity representing the linear relationship between each instrument signal, the similarity matrix for the entire instrument signal is

. .

Prediction method for monitoring the performance of power plant instrument using dependent variable similarity and kernel regression, characterized in that shown. The method according to claim 1, The fourth step, When you separately determine the Euclidean distance from the training data for the first test data,

. . .

Prediction method for power plant instrument performance monitoring using the dependent variable similarity and kernel regression, characterized in that obtained as follows. The method according to claim 1, The fifth step, In order to consider the importance between the instrument signals, the similarity matrix is weighted to the matrix of Euclidean distance, and

. .

About the trnth Train data set

. .

Prediction method for monitoring the performance of power plant instrument using dependent variable similarity and kernel regression, characterized in that performed as follows. The method according to claim 1, The fifth step, In order to take into account the importance between the instrument signals, the similarity matrix is weighted to the matrix of Euclidean distance, and the Euclidean distance matrix

Prediction method for power plant instrument performance monitoring using the dependent variable similarity and kernel regression, characterized in that obtained as follows. The method according to claim 1, The sixth step, When using the kernel function to find the weight for a given test data for each training data,

. .

Prediction method for power plant instrument performance monitoring using the dependent variable similarity and kernel regression, characterized in that obtained as follows. The method according to claim 1, The sixth step, When we use the kernel function to find the weight for a given test data for each training data, the weighting function

Prediction method for monitoring the performance of power plant instrument using dependent variable similarity and kernel regression method. The method according to claim 1, The seventh step, Given the predicted value of the first test data given by multiplying each weight by the training data,

And then from here

. .

Prediction method for monitoring the performance of power plant instrument using dependent variable similarity and kernel regression method.

Images