Evaluation method for on-line monitoring signal of nuclear power unit redundant instrument

Abstract

The invention belongs to the technical field of nuclear power station equipment management, and particularly relates to a nuclear power unit redundant instrument online monitoring signal evaluation method. Comprising the following steps of 1, collecting data; step 2, carrying out data normalization; step 3, data whitening; step 4, evaluation method; step 5, scale factor calculation; and 6, calculating a threshold value. The method has the beneficial effects that on the aspect of online monitoring signal processing, compared with a common mode that data is directly used as input of a model, data normalization processing is provided from the two aspects of data dimension and redundancy, so that the problem that scales of different physical parameters are different is solved; data whitening processing is provided, correlation between variables is removed, and independence of data between different dimensions is achieved.

Description

Technical Field

[0001] The invention belongs to the technical field of nuclear power plant equipment management, and in particular relates to an evaluation method for online monitoring signals of redundant instruments of a nuclear power unit. Background Art

[0002] Instruments in nuclear power plants will gradually drift after operation. If not calibrated in time, serious consequences will occur. At present, nuclear power plants stipulate that the calibration of instruments is a periodic active maintenance. During the overhaul window, the instruments in the work order tasks are isolated and directly calibrated on site or disassembled and brought to the laboratory for calibration. If the instrument is found to drift, it needs to be further calibrated.

[0003] The overly conservative calibration strategy under this traditional calibration method, due to the fixed calibration cycle and unnecessary calibration work, will bring potential risks to the safe operation of nuclear power plants, such as damage to instrumentation and equipment, causing calibration errors, increasing the workload during the refueling outage and even extending the outage time, and also increasing the possibility of radiation exposure to workers. In addition, in this time-based preventive maintenance method, the instrument is calibrated regularly, so even if the drift of the instrument exceeds the limit before the next calibration, it will continue to work, increasing the risk of safe and effective operation of nuclear power plants. Moreover, nuclear power plants have little experience in analyzing instrument data during the operation phase.

[0004] Based on the above situation, power plant engineers hope to find an online monitoring technology to monitor the performance of instrument channels during the operation phase, adjust the traditional regular calibration strategy to an online state-based calibration strategy, and optimize manual calibration of instruments that need to be calibrated during overhaul. Moreover, as the nuclear power industry gradually shifts to condition-based maintenance, online monitoring technology must be developed to better assist the development of nuclear power intelligence. Summary of the invention

[0005] The purpose of the present invention is to provide an evaluation method for online monitoring signals of redundant instruments of nuclear power units. By real-time monitoring signals of redundant instruments in nuclear power plants, the status of redundant instruments measuring the same parameter is evaluated. When some instrument signals show abnormalities, the abnormal status of the instruments can be discovered and judged in time, and timely feedback can be given to engineers, thereby improving the safety level of nuclear power plant operation.

[0006] The technical solution of the present invention is as follows: A method for evaluating online monitoring signals of redundant instruments of a nuclear power unit comprises the following steps:

[0007] Step 1: Collect data;

[0008] Step 2: Data normalization;

[0009] Step 3: Data whitening;

[0010] Step 4: Evaluation method;

[0011] Step 5: Calculation of scaling factor;

[0012] Step 6: Threshold calculation.

[0013] The step 1 is to identify the redundant instrument objects to be analyzed in the nuclear power plant, collect the instrument signal monitoring data for the time period to be analyzed, and organize and clean the data.

[0014] The step 4 is to record the observed data matrix after whitening as X, that is, X = X whitened , extract independent signal sources from the observation matrix X, the formula is as follows:

[0015]

[0016] Among them, S represents the independent component matrix, and A represents the unknown conversion coefficient;

[0017] By converting the above formula, the problem of solving the model is transformed into:

[0018] S=A -1 X=WX

[0019] That is, given the observed data matrix X, solve the mixing matrix W and independent components S;

[0020] First solve the mixing matrix W

[0021] According to the independence of the component S to be determined and combined with the principle of maximum likelihood estimation, the likelihood function is constructed as follows:

[0022]

[0023] Assuming its cumulative distribution function, the mathematical expression of the function is:

[0024]

[0025] Combining equations (1) and (2), taking logarithms and taking derivatives, we have:

[0026]

[0027] Using stochastic gradient optimization, we can solve W in an iterative way, and we have:

[0028]

[0029] Until W converges, the mixing matrix W is obtained.

[0030] Then solve the independent component S

[0031] After obtaining the mixing matrix W, the independent component S can be obtained according to the transformation S=WX.

[0032] The step 5 described above scales the solution back to its original unit. The calculation of the scaling factor α is described below.

[0033]

[0034] Where X is the data matrix of m redundant sensor signals, IC j is the jth independent component in the independent component S.

[0035] Next, the optimized mixing matrix is ​​calculated as follows:

[0036] W C =sign(α)·α·W

[0037] Wherein, sign(α) is a sign function, when α>0, the value is 1; when α=0, the value is 0; when α<0, the value is -1;

[0038] The final estimated values ​​of the evaluation method are as follows:

[0039] Y=W C X.

[0040] The step 6 is based on the step 5 to evaluate the instrument signal status, and the evaluation includes two parts: residual calculation and residual band setting.

[0041] The residual is calculated as follows:

[0042] Residual=YX (3)

[0043] The residual error of each instrument signal is determined by equation (3).

[0044] The residual band is set as follows:

[0045] The upper limit of the residual band is the mean of the n data points estimated + the standard deviation of the λ*n data points estimated, recorded as the upper limit value;

[0046] The lower limit of the residual band is: the estimated mean of n data points - λ*the estimated standard deviation of n data points, recorded as the lower limit value; the value of λ is 1, 2, 3...

[0047] The beneficial effects of the present invention are as follows: the evaluation method of online monitoring signals of redundant instruments of nuclear power units provided by the present invention provides an online monitoring technology for power plant engineers, realizes the monitoring and evaluation of the performance of redundant instrument channels during the operation phase, promotes the nuclear power industry to transform from the traditional regular preventive maintenance strategy to state-based predictive maintenance, and promotes the intelligent development of the nuclear power industry. The evaluation method of online monitoring signals of redundant instruments of nuclear power units provided by the present invention, regarding the processing of online monitoring signals, compared with the commonly used direct use of data as the input of the model, the present invention proposes a normalization process for the data from the two aspects of the dimension and redundancy of the data to eliminate the problem of different scales of different physical parameters; proposes a whitening process for the data to remove the correlation between variables and realize the independence of data between different dimensions. Moreover, the processed data is used as the input of the model, so that the convergence of the subsequent evaluation method is significantly improved. The present invention provides a method for evaluating online monitoring signals of redundant instruments of nuclear power units. Regarding the evaluation method, compared with the commonly used cross comparison to evaluate the status of redundant instrument online monitoring signals, the present invention starts from the perspective of independent component analysis, decomposes redundant multivariate signals into statistically independent components, and extracts independent signal sources from redundant signals to obtain an estimate of the most realistic information of the data. The present invention provides a method for evaluating online monitoring signals of redundant instruments of nuclear power units. Regarding the determination of instrument signal status, compared with the commonly used fixed thresholds in the alarm response procedures to discover and feedback abnormalities. Starting from the monitoring instrument itself, the present invention calculates the upper and lower limits of customized thresholds according to the trend performance of the data to achieve the sensitivity of the instrument signal status evaluation. Moreover, it can be clearly seen in the following embodiments that the threshold construction method of the present invention can timely monitor the occurrence of abnormalities and accurately locate the abnormal instrument compared with the fixed thresholds in the procedures. The design of the evaluation method of online monitoring signals of redundant instruments of nuclear power units of the present invention is novel and practical. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flow chart of an evaluation method for online monitoring signals of redundant instruments of a nuclear power unit provided by the present invention;

[0049] Figure 2 This is the time diagram of the online monitoring data of redundant instruments. DETAILED DESCRIPTION

[0050] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0051] A method for evaluating online monitoring signals of redundant instruments of a nuclear power unit comprises the following steps:

[0052] Step 1: Collect Data

[0053] Identify the redundant instrument objects to be analyzed in the nuclear power plant, collect the instrument signal monitoring data for the time period to be analyzed, and organize and clean the data to ensure the validity of the data.

[0054] Step 2: Data Normalization

[0055] Since the scales of different physical parameters of nuclear power plants are different, in the subsequent evaluation method, data normalization is first performed to eliminate the scale characteristics between the dimensions of the original data and improve the convergence speed of the model. The commonly used scaling intervals are [-1, +1] and [0, 1], but not limited to these two. Commonly used normalization methods include minimum-maximum normalization, zero-mean normalization, decimal calibration normalization, etc.

[0056] The mapping function formula of Min-Max Scaling is:

[0057]

[0058] Among them, x is the original value of the data, x max is the maximum value in the data set, x min is the minimum value in the data set, x * The new value after data mapping.

[0059] The mapping function formula for zero-mean normalization (Z-score) is:

[0060]

[0061] Among them, x is the original data, is the original data mean, δ is the original data standard deviation, x * is the new data after mapping. The data normalized by the Z-score method is unified into a new data set with a mean of 0 and a standard deviation of 1, which conforms to the standard normal distribution.

[0062] Step 3: Data whitening

[0063] The purpose of whitening is to reduce the redundancy of the original measurement data, and the whitened data has the following two characteristics compared to the original data:

[0064] (1) Decorrelation of different dimensions of data, that is, removing the correlation between variables;

[0065] (2) Each dimension of the data has the same unit variance;

[0066] In addition to the correlation between the observation signals of each instrument, the convergence performance of the method is significantly improved compared with the original data without whitening. There are two commonly used whitening methods:

[0067] PCA whitening

[0068] After the original data is whitened by PCA, the covariance matrix of the new data set is fixed to a unit matrix, with all diagonal elements being 1 and all off-diagonal elements being 0. That is, after whitening, the dimensions of the data are uncorrelated and the variance of each dimension is normalized to 1.

[0069] ●ZCA whitening

[0070] In order to make the whitened data as close as possible to the original data, the whitened data is not rotated (coordinate transformation) relative to the original space (coordinate system). The mapping function of whitening is:

[0071] X whitened =VX centered

[0072] V represents the eigenvalue matrix of the original data.

[0073] Step 4: Evaluation Methods

[0074] After processing the monitoring data of the nuclear power plant instrument in steps 2 and 3, the input conditions of the evaluation method are met. Next, the evaluation method is explained. The evaluation method is based on the idea of ​​independent principal component analysis. By decomposing redundant multivariate signals into statistically independent components, the independent signal source is extracted from the redundant signals to obtain the most realistic information of the data.

[0075] The observed data matrix after whitening is recorded as X, that is, X = X whitened The model of independent component analysis is to extract independent signal sources from the observation matrix X. The formula is as follows:

[0076]

[0077] Among them, S represents the independent component matrix and A represents the unknown conversion coefficient, also known as the mixing matrix.

[0078] By converting the above formula, the problem of solving the model is transformed into:

[0079] S=A -1 X=WX

[0080] That is, given the observed data matrix X, solve the mixing matrix W and independent component S.

[0081] First solve the mixing matrix W

[0082] According to the independence of the component S to be determined, combined with the principle of maximum likelihood estimation, the likelihood function is constructed as follows:

[0083]

[0084] Since we cannot directly make assumptions about the probability density function of S, we make assumptions about its cumulative distribution function. The mathematical expression of the function is:

[0085]

[0086] According to experience, the above sigmoid function can achieve good results in most problems.

[0087] Combining equations (1) and (2), taking logarithms and taking derivatives, we have:

[0088]

[0089] According to the above derivative expression of W, it is difficult to directly obtain the analytical expression of W. Therefore, we use stochastic gradient optimization to solve W in an iterative way, and we have:

[0090]

[0091] Until W converges, the mixing matrix W is obtained.

[0092] Then solve the independent component S

[0093] After obtaining the mixing matrix W, the independent component S can be obtained according to the transformation S=WX.

[0094] Step 5: Scale Factor Calculation

[0095] Since the mixing matrix W and independent component S obtained in step 4 have normalized dimensions, in order to compare with the original measured data, the solution needs to be scaled back to its original unit. The calculation of the scale factor α is described below.

[0096]

[0097] Where X is the data matrix of m redundant sensor signals, IC j is the jth independent component in the independent component S.

[0098] Next, the optimized mixing matrix is ​​calculated as follows:

[0099] W C =sign(α)·α·W

[0100] Wherein, sign(α) is a sign function, and when α>0, the value is 1; when α=0, the value is 0; when α<0, the value is -1.

[0101] Therefore, the final estimated value of the evaluation method is as follows:

[0102] Y=W C X

[0103] Step 6: Threshold calculation

[0104] Based on step 5, the instrument signal status is evaluated. The evaluation contains two parts: residual calculation and residual band setting.

[0105] ①

[0106] Residual=YX (3)

[0107] The residual error of each instrument signal is determined by equation (3).

[0108] ②

[0109] There are many ways to set the residual band. Based on engineering experience, the following is a commonly used and effective way to set the residual band:

[0110] The upper limit of the residual band is the mean of the n data points estimated + the standard deviation of the λ*n data points estimated, recorded as the upper limit value;

[0111] The lower limit of the residual band is: the mean estimated by n data points - the standard deviation estimated by λ*n data points, recorded as the lower limit.

[0112] The value of λ can be 1, 2, 3, etc. Generally, λ is 2.

[0113] Embodiment 1:

[0114] The specific application of the above method is introduced by taking the redundant channel instrument of a nuclear power plant as an example.

[0115] The data comes from the power plant's data acquisition system, with a sampling frequency of 2 minutes and a data length of 1000 data points. The selected instrument series redundancy is 3, which are recorded as signal 1, signal 2 and signal 3. In order to visualize the observation, the data is organized as Figure 2 shown.

[0116] Combined with the actual supplementary explanation of the power plant, according to the provisions of the power plant alarm response regulations, on the same header, if the deviation of any instrument from its middle value exceeds 0.1Mpa, an alarm will be triggered. The treatment measures to be taken are to contact the instrumentation and control personnel for inspection and, if necessary, calibrate the corresponding transmitter. Figure 2 Calculation and verification of the data showed that the maximum deviation between each instrument and its median value was 0.0648 MPa, which was lower than the triggering requirements in the alarm response procedures in the operation manual and did not trigger an alarm.

[0117] Using the above evaluation method to process the data, from the calculation results of the residual and residual band (λ=2), signal 2 is clearly identified as abnormal drift, and signals 1 and 3 are within the residual band range, and no drift is identified. Figure 2 In the monitoring data from the data acquisition system, signal 2 behaves abnormally, and the results are consistent; moreover, the abnormal changes in the signal can be detected in advance before the alarm is triggered.

[0118] In addition, the settings of different values ​​of the residual band λ are summarized in Table 1. It can be observed from the table that the anomaly of signal 2 can be clearly identified, and the result is consistent with the anomaly judgment of the original monitoring data; moreover, the abnormal change of the signal at this time did not trigger the power plant alarm, which further illustrates that the evaluation method has the characteristics of early detection of signal anomalies, which can be fed back to the power plant engineers as soon as possible, and can effectively improve the safety of the power plant.

[0119] Table 1 Redundant instrument signal status evaluation results

[0120]

Claims

1. A method for evaluating online monitoring signals of redundant instruments of nuclear power units, characterized in that: The steps include: Step 1: Collect data; Step 2: Data normalization; Step 3: Data whitening; Step 4: Evaluation method; Step 5: Calculation of scaling factor; Step 6: Threshold calculation.

2. The method for evaluating online monitoring signals of redundant instruments of a nuclear power unit according to claim 1, characterized in that: The step 1 is to identify the redundant instrument objects to be analyzed in the nuclear power plant, collect the instrument signal monitoring data for the time period to be analyzed, and organize and clean the data.

3. The method for evaluating online monitoring signals of redundant instruments of a nuclear power unit according to claim 1, characterized in that: The step 4 is to record the observed data matrix after whitening as X, that is, X = X denetiwh , extract independent signal sources from the observation matrix X, the formula is as follows: Among them, S represents the independent component matrix, and A represents the unknown conversion coefficient; By converting the above formula, the problem of solving the model is transformed into: S=A -1 X=WX That is, given the observed data matrix X, solve the mixing matrix W and independent component S.

4. A method for evaluating online monitoring signals of redundant instruments of nuclear power units as claimed in claim 3, characterized in that: The process of solving the mixing matrix W and the independent component S is as follows: First solve the mixing matrix W According to the independence of the component S to be determined and combined with the principle of maximum likelihood estimation, the likelihood function is constructed as follows: Assuming its cumulative distribution function, the mathematical expression of the function is: Combining equations (1) and (2), taking logarithms and taking derivatives, we have: Using stochastic gradient optimization, we can solve W in an iterative way, and we have: Until W converges, the mixing matrix W is obtained; After solving the independent component S and obtaining the mixing matrix W, the independent component S can be obtained according to the transformation S=WX.

5. The method for evaluating online monitoring signals of redundant instruments of a nuclear power unit according to claim 1, characterized in that: Step 5 scales the solution back to its original units. The calculation of the scaling factor α is as follows: Where X is the data matrix of m redundant sensor signals, IC j is the jth independent component in the independent component S; Next, the optimized mixing matrix is ​​calculated as follows: W C =sign(α)·α·W Wherein, sign(α) is a sign function, when α>0, the value is 1; when α=0, the value is 0; when α<0, the value is -1; The final estimated values ​​of the evaluation method are as follows: Y=W C X。 6. The method for evaluating online monitoring signals of redundant instruments of a nuclear power unit according to claim 1, characterized in that: The step 6 is based on the step 5 to evaluate the instrument signal status, and the evaluation includes two parts: residual calculation and residual band setting.

7. A method for evaluating online monitoring signals of redundant instruments of nuclear power units as claimed in claim 6, characterized in that: The residual is calculated as follows: Residual=YX (3) The residual error of each instrument signal is determined by equation (3).

8. A method for evaluating online monitoring signals of redundant instruments of nuclear power units as claimed in claim 6, characterized in that: The residual band is set as follows: The upper limit of the residual band is the mean of the n data points estimated + the standard deviation of the λ*n data points estimated, recorded as the upper limit value; The lower limit of the residual band is: the estimated mean of n data points - λ*the estimated standard deviation of n data points, recorded as the lower limit value; the value of λ is 1, 2, 3...

Images