Current transformer error on-line detection method and system based on data quadratic correction

Abstract

The application discloses a current transformer error state online detection method and system based on data secondary correction, and the method comprises the following steps: collecting historical data and real-time test data of a current transformer to form current data; performing primary correction on the test data to obtain primary correction data of the current data; performing secondary correction on the primary correction data to obtain secondary correction data of the current data; calculating Q statistics by using the secondary correction data, comparing the Q statistics with a statistics control limit Q c , and determining whether the error state of the current transformer is abnormal; the application has the advantages that the error state of the current transformer can be detected online in the case that the unbalance degree fluctuates greatly.

Description

Technical Field

[0001] This invention relates to the field of current transformer error detection, and specifically to an online error detection method and system for current transformers based on secondary data correction. Background Technology

[0002] A current transformer is a sensing device used to detect large current signals in a power system. Its working principle is to proportionally convert the large current signal on the primary side into a small current signal on the secondary side, providing accurate primary current information for measurement, control, and protection equipment. As a signal sensing device, error is a key indicator for evaluating the performance of a current transformer. Under the influence of various complex factors such as harmonics, DC bias, residual magnetism, leakage current, and temperature, current transformers are at risk of error degradation, affecting the safe, stable, and reliable operation of the power system. Therefore, it is necessary to promptly monitor the error status of current transformers in operation.

[0003] Traditional error detection methods for current transformers involve periodically comparing them with standard current transformers under offline or live conditions. However, this method lacks real-time capability and cannot promptly detect the error status of the current transformer. Furthermore, on-site error verification based on this method suffers from numerous problems, including high workload and safety risks. Therefore, there is a need to explore online error detection methods for current transformers that operate independently of standard current transformers.

[0004] Chinese Patent Publication No. CN107255792A discloses a method and system for online monitoring of errors in electronic voltage transformers. The method includes: collecting measurement data from a three-phase electronic voltage transformer under operating conditions, establishing a residual model to calculate its statistical characteristics, and comparing these with statistical control limits to determine whether the error state of the electronic voltage transformer is normal. This online monitoring method for the error state of electronic voltage transformers provides a way to achieve real-time online monitoring of the error state of electronic voltage transformers, eliminating the need for periodic testing with traditional standards. However, this method struggles to analyze current data with relatively large unbalance fluctuations, making it difficult to achieve online detection of the error state of current transformers. Summary of the Invention

[0005] The technical problem to be solved by the present invention is that the existing online detection method for current transformer error is difficult to analyze current data with relatively large unbalance fluctuations, thus making it difficult to realize online detection of the error state of the current transformer.

[0006] This invention solves the above-mentioned technical problems through the following technical means: an online error detection method for current transformers based on secondary data correction, comprising:

[0007] S1. Collect historical data and real-time test data from the current transformer to form current data;

[0008] S2. Calculate the three-phase unbalance sequence based on the current data, calculate the current standard deviation using the three-phase unbalance sequence and compare it with the data correction standard. If the current standard deviation is greater than the data correction standard, then the current data is corrected once to obtain the corrected data.

[0009] S3. Calculate the historical current deviation of each phase of the historical data, perform empirical wavelet transform on it, and establish an autoregressive moving average prediction model to predict the theoretical predicted value of the historical current deviation of each phase; invert the theoretical predicted value of the historical current deviation of each phase to obtain the theoretical predicted value of the historical current data; perform a second correction on the first correction data of the current data based on the theoretical predicted value of the historical current data to obtain the second correction data.

[0010] S4. Calculate the Q statistic using the second-corrected data, and compare the Q statistic with the control limit Q. c By comparing the data, it can be determined whether the error state of the current transformer is abnormal.

[0011] Furthermore, S2 includes:

[0012] S2.1 The maximum value obtained by taking the absolute difference between the A, B, and C phase data of the current data and the three-phase current average value of the current data, and dividing the maximum value by the three-phase current average value of the current data, yields the three-phase unbalance sequence CUF.

[0013] S2.2 Calculate the standard deviation of the current using the standard deviation of the three-phase unbalance sequence CUF and the average value of the three-phase unbalance sequence CUF;

[0014] S2.3 Determine the data correction standards;

[0015] S2.4. Compare the current standard deviation with the data correction standard. If the current standard deviation is less than or equal to the data correction standard, no correction is needed for the current data. If the current standard deviation is greater than the data correction standard, the current data is corrected once to obtain corrected data.

[0016] Furthermore, S2.4 includes:

[0017] S2.4.1 If the standard deviation of the current is less than the data correction standard, it is determined that the current data does not need to be corrected, and S2.4.3 is executed; if the standard deviation of the current is greater than the data correction standard, it is determined that the current data needs to be corrected, and S2.4.2 is executed.

[0018] S2.4.2 Perform an affine transformation on the current data, use the result of the affine transformation as the current data, and return to S2.1; the affine transformation is to multiply the current data by the linear transformation parameter and then add the translation transformation parameter;

[0019] S2.4.3 Output the current current data as a correction of the current data.

[0020] Furthermore, S3 includes:

[0021] S3.1 Subtract the average three-phase current of the historical data from the historical data for each phase data, and then divide by the average three-phase current of the historical data to obtain the historical current deviation of each phase of the historical data;

[0022] S3.2. The historical current deviation of each phase in the historical data is obtained by empirical wavelet transform to obtain the modal components of the historical current deviation of each phase.

[0023] S3.3 Establish an autoregressive moving average prediction model, use the historical current deviation modal components of each phase as the past observations of the autoregressive moving average prediction model, and predict the theoretical values ​​of the historical current deviation of each phase.

[0024] S3.4. Perform inversion calculation on the theoretical predicted value of the historical current deviation of each phase. The inversion calculation formula is the theoretical predicted value of the historical current data minus the average value of the three-phase current of the historical data, and then divided by the average value of the three-phase current of the historical data. This equals the theoretical predicted value of the historical current deviation of each phase. Substitute the theoretical predicted value of the historical current deviation of each phase into the inversion calculation formula to obtain the theoretical predicted value of the historical current data.

[0025] S3.5. Based on the theoretical prediction value of historical current data, the first correction data of the current data is corrected a second time to obtain the second correction data.

[0026] Furthermore, S3.5 includes:

[0027] S3.5.1. The theoretical predicted value of the historical current data is divided by the test data, and the three-phase average of the quotient is calculated as the mean value characteristic of the quotient.

[0028] S3.5.2. Multiply the mean characteristic of the quotient with the first-correction data to obtain the second-correction data of the current data.

[0029] Furthermore, S4 includes:

[0030] S4.1. Standardize the data after secondary correction to obtain a standardized matrix;

[0031] S4.2 Decompose the standardized matrix using principal component analysis, resulting in a principal component model and a residual model. The principal component model is the product of the principal score matrix and the transpose of the principal loading matrix, while the residual model is the product of the residual score matrix and the transpose of the residual loading matrix.

[0032] S4.3 Perform singular value decomposition on the covariance matrix of the standardized matrix to obtain multiple eigenvalues. Sort the eigenvalues ​​from largest to smallest and calculate the cumulative percentage of the first p1 eigenvalues. Make the cumulative percentage of the first p1 eigenvalues ​​just greater than the set limit to obtain the corresponding p1. Then construct the principal component loading matrix based on the first p1 eigenvalues ​​and construct the residual loading matrix based on the remaining eigenvalues.

[0033] S4.4 Calculate the projection of the normalized matrix onto the residual loading matrix to obtain the residual score matrix;

[0034] S4.5. The Q statistic of the standardized matrix is ​​obtained by multiplying the transpose of the residual score matrix, the residual loading matrix, and the transpose of the standardized matrix; if Q < Q c If Q > Q, then it is determined that the three-phase current transformer is operating normally at this time. c If so, it is determined that the error state of the three-phase current transformer is in an abnormal state.

[0035] Furthermore, the online error detection method for current transformers based on secondary data correction also includes S5 for model training, which includes:

[0036] After testing b test data points in the current round, calculate the Q statistic for each of the b test data points in the current round that exceeds the statistic control limit. c If m < 30%b, discard the earliest b test data of the current data, i.e. the b test data of the first round, and update the test data of the current data for the next round, and continue to execute S2; if m ≥ 30%b, discard the b test data of the current round, and update the test data of the current data for the next round, and continue to execute S2.

[0037] This invention also provides an online error detection system for current transformers based on secondary data correction, comprising:

[0038] The data acquisition module is used to collect historical data and real-time test data from the current transformer to form current data.

[0039] The first correction module is used to calculate the three-phase unbalance sequence based on the current data, calculate the current standard deviation using the three-phase unbalance sequence and compare it with the data correction standard. If the current standard deviation is greater than the data correction standard, the current data is corrected once to obtain the corrected data.

[0040] The second correction module is used to calculate the historical current deviation of each phase of the historical data, perform empirical wavelet transform on it and establish an autoregressive moving average prediction model to predict the theoretical predicted value of the historical current deviation of each phase; inversely calculate the theoretical predicted value of the historical current data based on the theoretical predicted value of the historical current deviation of each phase; and perform a second correction on the first correction data of the current data based on the theoretical predicted value of the historical current data to obtain the second correction data.

[0041] The online detection module is used to calculate the Q statistic using the second-corrected data, and then compares the Q statistic with the control limit Q. c By comparing the data, it can be determined whether the error state of the current transformer is abnormal.

[0042] Furthermore, the first correction module is also used for:

[0043] S2.1 The maximum value obtained by taking the absolute difference between the A, B, and C phase data of the current data and the three-phase current average value of the current data, and dividing the maximum value by the three-phase current average value of the current data, yields the three-phase unbalance sequence CUF.

[0044] S2.2 Calculate the standard deviation of the current using the standard deviation of the three-phase unbalance sequence CUF and the average value of the three-phase unbalance sequence CUF;

[0045] S2.3 Determine the data correction standards;

[0046] S2.4. Compare the current standard deviation with the data correction standard. If the current standard deviation is less than or equal to the data correction standard, no correction is needed for the current data. If the current standard deviation is greater than the data correction standard, the current data is corrected once to obtain corrected data.

[0047] Furthermore, S2.4 includes:

[0048] S2.4.1 If the standard deviation of the current is less than the data correction standard, it is determined that the current data does not need to be corrected, and S2.4.3 is executed; if the standard deviation of the current is greater than the data correction standard, it is determined that the current data needs to be corrected, and S2.4.2 is executed.

[0049] S2.4.2 Perform an affine transformation on the current data, use the result of the affine transformation as the current data, and return to S2.1; the affine transformation is to multiply the current data by the linear transformation parameter and then add the translation transformation parameter;

[0050] S2.4.3 Output the current current data as a correction of the current data.

[0051] Furthermore, the second correction module is also used for:

[0052] S3.1 Subtract the average three-phase current of the historical data from the historical data for each phase data, and then divide by the average three-phase current of the historical data to obtain the historical current deviation of each phase of the historical data;

[0053] S3.2. The historical current deviation of each phase in the historical data is obtained by empirical wavelet transform to obtain the modal components of the historical current deviation of each phase.

[0054] S3.3 Establish an autoregressive moving average prediction model, use the historical current deviation modal components of each phase as the past observations of the autoregressive moving average prediction model, and predict the theoretical values ​​of the historical current deviation of each phase.

[0055] S3.4. Perform inversion calculation on the theoretical predicted value of the historical current deviation of each phase. The inversion calculation formula is the theoretical predicted value of the historical current data minus the average value of the three-phase current of the historical data, and then divided by the average value of the three-phase current of the historical data. This equals the theoretical predicted value of the historical current deviation of each phase. Substitute the theoretical predicted value of the historical current deviation of each phase into the inversion calculation formula to obtain the theoretical predicted value of the historical current data.

[0056] S3.5. Based on the theoretical prediction value of historical current data, the first correction data of the current data is corrected a second time to obtain the second correction data.

[0057] Furthermore, S3.5 includes:

[0058] S3.5.1. The theoretical predicted value of the historical current data is divided by the test data, and the three-phase average of the quotient is calculated as the mean value characteristic of the quotient.

[0059] S3.5.2. Multiply the mean characteristic of the quotient with the first-correction data to obtain the second-correction data of the current data.

[0060] Furthermore, the online detection module is also used for:

[0061] S4.1. Standardize the data after secondary correction to obtain a standardized matrix;

[0062] S4.2 Decompose the standardized matrix using principal component analysis, resulting in a principal component model and a residual model. The principal component model is the product of the principal score matrix and the transpose of the principal loading matrix, while the residual model is the product of the residual score matrix and the transpose of the residual loading matrix.

[0063] S4.3 Perform singular value decomposition on the covariance matrix of the standardized matrix to obtain multiple eigenvalues. Sort the eigenvalues ​​from largest to smallest and calculate the cumulative percentage of the first p1 eigenvalues. Make the cumulative percentage of the first p1 eigenvalues ​​just greater than the set limit to obtain the corresponding p1. Then construct the principal component loading matrix based on the first p1 eigenvalues ​​and construct the residual loading matrix based on the remaining eigenvalues.

[0064] S4.4 Calculate the projection of the normalized matrix onto the residual loading matrix to obtain the residual score matrix;

[0065] S4.5. The Q statistic of the standardized matrix is ​​obtained by multiplying the transpose of the residual score matrix, the residual loading matrix, and the transpose of the standardized matrix; if Q < Q c If Q > Q, then it is determined that the three-phase current transformer is operating normally at this time. c If so, it is determined that the error state of the three-phase current transformer is in an abnormal state.

[0066] Furthermore, the online error detection system for current transformers based on secondary data correction also includes a training module for model training. The training module is used for:

[0067] After testing b test data points in the current round, calculate the Q statistic for each of the b test data points in the current round that exceeds the statistic control limit. c If m < 30%b, discard b test data points from the first round of current data, update the next round of current data test data, and continue executing S2; if m ≥ 30%b, discard b test data points from the current round, update the next round of current data test data b test data, and continue executing the first correction module.

[0068] The advantages of this invention are:

[0069] (1) This invention calculates the three-phase unbalance sequence based on current data, calculates the current standard deviation using the three-phase unbalance sequence and compares it with the data correction standard. If the current standard deviation is greater than the data correction standard, the current data is corrected once to obtain the first-corrected data. This enables the analysis and correction of current data with relatively large unbalance fluctuations. Furthermore, the first-corrected data of the current data is corrected a second time based on the theoretical prediction value of historical current data. Then, principal component analysis is performed on the second-corrected data to establish a residual model, and the Q statistic is calculated to determine the degree of deviation of non-principal components. The calculated Q statistic considers the influence of three-phase unbalance and enhances the error characteristics in the data through the second correction, thereby ultimately achieving accurate online detection of the error state of the current transformer.

[0070] (2) The online detection method for current transformer error status based on data secondary correction provided by the present invention analyzes the error status of current transformer based on the measurement data of the current transformer itself. It does not require the use of standard current transformers and has the advantages of low detection cost and strong real-time performance compared with traditional current transformer error status detection methods.

[0071] (3) The online detection method for current transformer error status based on secondary data correction provided by the present invention detects the error status of the current transformer online based on the measurement data of the current transformer itself. Compared with the traditional current transformer error status detection method, it can effectively reduce the offline testing workload and test safety hazards of power grid operation and maintenance personnel. Attached Figure Description

[0072] Figure 1 This is a flowchart of the online detection method for current transformer error status based on secondary data correction disclosed in an embodiment of the present invention.

[0073] Figure 2 This is a schematic diagram of the online detection system for current transformer error status based on secondary data correction disclosed in an embodiment of the present invention. Detailed Implementation

[0074] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] This invention provides an online error state detection method for current transformers based on secondary data correction. The detection target is a three-phase current transformer within a specific interval. The specific approach is as follows: First, affine transformation is used to preliminarily correct the synchronously acquired node three-phase current data, establishing a new comparison standard with correlation constraints. Second, empirical wavelet transform-autoregressive moving average is used to extract residual components containing error information, and the data is further corrected by combining the daily periodic characteristics of the system current to enhance its error characteristics. Finally, an improved moving window principal component analysis method is used to dynamically evaluate the second-corrected data to achieve long-term online detection of current transformer measurement errors. This invention can effectively reduce the cost of current transformer error detection and the safety hazards for personnel, and also provides a new approach to online detection methods for current transformers. The flowchart of the online error state detection method for current transformers based on secondary data correction proposed in this invention is as follows: Figure 1 As shown, the specific steps include the following:

[0076] S1: Obtain a historical data points measured for each of the three phases of the three-phase current transformer. And collect b test data points from the three phases of the three-phase current transformer in real time. Historical data and test data together constitute current data. Where x represents phase A, phase B, or phase C of the three-phase current transformer.

[0077] S2: Since the fluctuation of current imbalance in the same line is much greater than the fluctuation of voltage imbalance, the current data can be corrected once. Based on the corrected data, a new online detection comparison standard is established to minimize the fluctuation of imbalance, and the corrected current data is obtained. Specifically, this includes the following sub-steps:

[0078] S2.1: Based on the current data I measured by the three-phase current transformers respectively x Calculate the three-phase unbalance sequence

[0079]

[0080] Among them, I A I B I C These are the current data I. x The A, B, and C phase currents in the circuit. ave isI A I B I C Average value. S2.2: Calculate the standard deviation of the current based on the three-phase unbalance calculation results using the current data.

[0081]

[0082] Where i refers to the i-th data point, CUF ave It refers to the average value of the three-phase unbalance series CUF.

[0083] S2.3: Determine the data correction standard RS based on the time-domain characteristics of the standard deviation of the three-phase voltage data imbalance under stable operation of the power system; this data correction standard is a data set based on the short-term invariance of the three-phase voltage data in the actual power system, and can be adjusted as needed in practical applications.

[0084] S2.4: Compare the current standard deviation S CUF The result is compared with the data correction standard RS to determine whether data correction is needed. Specifically, this includes the following sub-steps:

[0085] S2.4.1: If the current standard deviation S CUF If the current standard deviation is less than the data correction standard RS, then it is determined that the current data does not require data correction, and S2.4.3 is executed; if the current standard deviation S... CUF If the current data is greater than the data correction standard RS, then it is determined that the current data needs to be corrected, and S2.4.2 is executed;

[0086] S2.4.2: Perform an affine transformation on the current data affine transformation result As current data I x Return to S2.1; where A1 represents the linear transformation parameters, which can be the identity matrix E; b1 represents the translation transformation parameters, which can be b1 = [1,1,…,1]. T ;

[0087] S2.4.3: Output the current current data as a first-order correction data I of the current data. x aff1 ;

[0088] S3: S2 may weaken the first correction data I x aff1 The original error information in the data needs to be combined with the daily cycle time-domain characteristics existing under normal operation of the power system to perform secondary correction on the current data and enhance the error characteristics in the data, thereby obtaining secondary corrected current data; specifically, it includes the following sub-steps:

[0089] S3.1: Calculate the historical current deviation of each phase of the historical data. This represents the average three-phase current from historical data. The x-phase data represents historical data;

[0090] S3.2: In practice, in order to eliminate the historical current deviation CUF of each phase... x his The impact of nonstationarity on the S3 results requires analysis of the historical current deviation CUF for each phase. x his Preprocessing is performed to obtain the historical current deviation mode components for each phase. Preprocessing methods can include empirical wavelet transform, which is an existing technology. Its technical principle is as follows:

[0091] I0(t)=W x (0,t)*φ1(t),

[0092] Wherein, W x (0,t) are approximation coefficients; W x (n,t) are detail coefficients; φ1(t) is the empirical scaling function; The function is the empirical wavelet function; I(t) is the input phase current data. x The time series.

[0093] In this embodiment, the empirical wavelet transform can be adopted from the empirical wavelet transform method described in the literature "Gilles, J. Empirical wavelet transform[J].IEEE Transactions on Signal Processing, 2013, 61(16):3999–4010."

[0094] S3.3: Establish an autoregressive moving average prediction model, using the historical current deviation modal components of each phase as past observations for the autoregressive moving average prediction model to predict future values. In other words, use the autoregressive moving average prediction model to obtain the theoretical predicted values ​​of the historical current deviations for each phase. The mathematical model of the existing autoregressive moving average model is as follows:

[0095]

[0096] Where p is the order of the autoregressive model; The undetermined coefficients of the model; ε t θ is the error; q is the order of the moving average model; θ j (j=1,2,…,q) are the undetermined coefficients of the model.

[0097] In this embodiment, the autoregressive moving average prediction model can be adopted from the literature "Box, GEP, Jenkins, GM, et al. Time Series Analysis: Forecasting and Control, 5th edition [M]. Hoboken: Wiley, 2015."

[0098] S3.4: Theoretical prediction of the historical current deviation for each phase Inversion calculations are performed, using the same formula as in S3.1, to obtain the theoretical predicted values ​​from the historical current data.

[0099] S3.5: First correction data I of the current data based on the theoretical prediction value of historical current data. x aff1 Perform a second correction; specifically including the following sub-steps:

[0100] S3.5.1: Take the quotient of the theoretical predicted value of the historical current data and the test data, and calculate the mean characteristic of the quotient, expressed by the formula:

[0101] S3.5.2: Multiply the mean characteristic of the quotient value with the first-correction data to obtain the second-correction data I of the current data.x aff2 .

[0102] S4: Dynamically and online detect the error state of the current transformer; specifically including the following sub-steps:

[0103] S4.1: The matrix form of the second-order corrected data is as follows Let X be the second-order corrected data matrix. Standardizing the second-order corrected data matrix yields the standardized matrix X = [X...]. A X B X C ] (a+b)×3 ;

[0104] S4.2: The standardized matrix is ​​decomposed using Principal Component Analysis (PCA) according to the following formula:

[0105]

[0106] Among them, t i For the i-th score vector, p i T For the i-th load vector, For the principal component model of the standardized matrix, E = T e P e T For the residual model of the standardized matrix, T is the principal component score matrix of the standardized matrix, and P is the residual model of the standardized matrix. T T is the transpose of the principal component loading matrix of the normalized matrix. e P is the residual score matrix of the standardized matrix. e T This is the transpose of the residual load matrix of the normalized matrix.

[0107] S4.3: Perform singular value decomposition on the covariance matrix R of the standardized matrix to obtain n eigenvalues. Sort the eigenvalues ​​from largest to smallest and calculate the cumulative percentage of the top p1 eigenvalues. Where, λ j Let be the j-th eigenvalue, and n be the total number of eigenvalues. CPV(p1) can typically be capped at 85%, ensuring that CPV(p1) is greater than 85% when the first p1 eigenvalues ​​are substituted into the calculation. Once the CPV(p1) cap is determined, the corresponding p1 can be calculated. Then, the principal component loading matrix P is constructed based on the first p1 eigenvalues, and the residual loading matrix P is constructed using the remaining (n-p1) eigenvalues. e The formula for singular value decomposition is R = X. T X / (n-1)=[PP e ]Λ[PP e ] TWhere Λ = diag(λ1, λ2, λ3) are the eigenvalues ​​of the covariance matrix R, λ1 ≥ λ2 ≥ λ3, [PP e ] is the feature matrix formed by the eigenvectors corresponding to the eigenvalues ​​λ1, λ2, and λ3.

[0108] S4.4: Residual score matrix T of the standardized matrix e The normalized matrix X is in the residual loading matrix P e The projection on, that is, T e =XP e The residual loading matrix P was obtained through S4.3. e Thus, T is substituted. e =XP e The residual score matrix T can be obtained. e .

[0109] S4.5: According to the formula Q = (XP) e P e T (XP) e P e T ) T =XP e P e T X T Calculate the Q-statistic of the standardized matrix and calculate the control limits of the statistic. And determine the operating status of the three-phase current transformer at this time; where C α The parameter represents the critical value of a normal distribution at a given significance level α. θ i denoted by i to the power of the non-principal eigenvalue; n represents the total number of eigenvalues ​​after singular value decomposition; k represents the number of principal components retained; j represents the j-th eigenvalue, which is a non-principal eigenvalue. Let i represent the i-th power of the j-th eigenvalue.

[0110] If Q<Q c If so, it can be determined that the three-phase current transformer is operating normally at this time;

[0111] If Q>Q c If so, it is determined that the error state of the three-phase current transformer is in an abnormal state.

[0112] S5. After detecting b test data in the current round, calculate whether the statistic Q exceeds the statistic control limit Q. c The number of times m is determined, and the update mode of the online detection model is judged:

[0113] If m < 30%b, then discard the earliest b test data of the current data, which is the b test data of the first round, and update the test data of the current data for the next round, and continue to execute S2; if m < 30%b, it means that the three-phase current transformer is basically in normal condition. At this time, in order to prevent the online detection model from getting trapped in a local optimum, the earliest test data is removed and new test data is added to improve the training accuracy of the model.

[0114] If m ≥ 30%b, discard b test data points from the current round and update the next b test data points for the current data, then continue executing S2. m ≥ 30%b indicates that the error state of the three-phase current transformer is abnormal. In this case, to prevent detection anomalies caused by incorrect test data, the test data from the current round is discarded, the test data is updated, and the model's detection accuracy is improved.

[0115] Through the above technical solutions, the online detection method for current transformer error status based on secondary data correction provided by this invention analyzes the error status of the current transformer based on its own measurement data. It does not require the use of a standard current transformer and has the advantages of low detection cost and strong real-time performance compared with traditional current transformer error status detection methods. Compared with traditional current transformer error status detection methods, it can effectively reduce the offline testing workload and testing safety hazards of power grid operation and maintenance personnel.

[0116] Example 2

[0117] Based on Embodiment 1, Embodiment 2 of the present invention also provides an online error detection system for current transformers based on secondary data correction. The detection target is a three-phase current transformer within a specific interval, and its system configuration is as follows: Figure 2 As shown, the system includes an interconnected data correction module and an online detection module. The data correction module comprises a first correction module and a second correction module. The data correction module corrects three-phase current data, establishes a novel comparison standard suitable for online detection of current transformer error states based on the operating characteristics of the power system and data transformation, and sends the corrected data to the online detection module. The online detection module detects the error states of current transformers online and updates the current acquisition data using the detection model, then sends the updated current data to the data correction module. The system also includes a data acquisition module and a training module. Figure 2 (Not shown in the diagram) The data acquisition module is connected to the first correction module, the first correction module is connected to the second correction module, the second correction module is connected to the online detection module, and the training module trains the online detection model composed of the data acquisition module, the data correction module, and the online detection module. The online detection system specifically includes:

[0118] The data acquisition module is used to collect historical data and real-time test data from the current transformer to form current data.

[0119] The first correction module is used to calculate the three-phase unbalance sequence based on the current data, calculate the current standard deviation using the three-phase unbalance sequence and compare it with the data correction standard. If the current standard deviation is greater than the data correction standard, the current data is corrected once to obtain the corrected data.

[0120] The second correction module is used to calculate the historical current deviation of each phase of the historical data, perform empirical wavelet transform on it and establish an autoregressive moving average prediction model to predict the theoretical predicted value of the historical current deviation of each phase; inversely calculate the theoretical predicted value of the historical current data based on the theoretical predicted value of the historical current deviation of each phase; and perform a second correction on the first correction data of the current data based on the theoretical predicted value of the historical current data to obtain the second correction data.

[0121] The online detection module is used to calculate the Q statistic using the second-corrected data, and then compares the Q statistic with the control limit Q. c By comparing the data, it can be determined whether the error state of the current transformer is abnormal.

[0122] Specifically, the first correction module is also used for:

[0123] S2.1 The maximum value obtained by taking the absolute difference between the A, B, and C phase data of the current data and the three-phase current average value of the current data, and dividing the maximum value by the three-phase current average value of the current data, yields the three-phase unbalance sequence CUF.

[0124] S2.2 Calculate the standard deviation of the current using the standard deviation of the three-phase unbalance sequence CUF and the average value of the three-phase unbalance sequence CUF;

[0125] S2.3 Determine the data correction standards;

[0126] S2.4. Compare the current standard deviation with the data correction standard. If the current standard deviation is less than or equal to the data correction standard, no correction is needed for the current data. If the current standard deviation is greater than the data correction standard, the current data is corrected once to obtain corrected data.

[0127] More specifically, S2.4 includes:

[0128] S2.4.1 If the standard deviation of the current is less than the data correction standard, it is determined that the current data does not need to be corrected, and S2.4.3 is executed; if the standard deviation of the current is greater than the data correction standard, it is determined that the current data needs to be corrected, and S2.4.2 is executed.

[0129] S2.4.2 Perform an affine transformation on the current data, use the result of the affine transformation as the current data, and return to S2.1; the affine transformation is to multiply the current data by the linear transformation parameter and then add the translation transformation parameter;

[0130] S2.4.3 Output the current current data as a correction of the current data.

[0131] Specifically, the second correction module is also used for:

[0132] S3.1 Subtract the average three-phase current of the historical data from the historical data for each phase data, and then divide by the average three-phase current of the historical data to obtain the historical current deviation of each phase of the historical data;

[0133] S3.2. The historical current deviation of each phase in the historical data is obtained by empirical wavelet transform to obtain the modal components of the historical current deviation of each phase.

[0134] S3.3 Establish an autoregressive moving average prediction model, use the historical current deviation modal components of each phase as the past observations of the autoregressive moving average prediction model, and predict the theoretical values ​​of the historical current deviation of each phase.

[0135] S3.4. Perform inversion calculation on the theoretical predicted value of the historical current deviation of each phase. The inversion calculation formula is the theoretical predicted value of the historical current data minus the average value of the three-phase current of the historical data, and then divided by the average value of the three-phase current of the historical data. This equals the theoretical predicted value of the historical current deviation of each phase. Substitute the theoretical predicted value of the historical current deviation of each phase into the inversion calculation formula to obtain the theoretical predicted value of the historical current data.

[0136] S3.5. Based on the theoretical prediction value of historical current data, the first correction data of the current data is corrected a second time to obtain the second correction data.

[0137] More specifically, S3.5 includes:

[0138] S3.5.1. The theoretical predicted value of the historical current data is divided by the test data, and the three-phase average of the quotient is calculated as the mean value characteristic of the quotient.

[0139] S3.5.2. Multiply the mean characteristic of the quotient with the first-correction data to obtain the second-correction data of the current data.

[0140] Specifically, the online detection module is also used for:

[0141] S4.1. Standardize the data after secondary correction to obtain a standardized matrix;

[0142] S4.2 Decompose the standardized matrix using principal component analysis, resulting in a principal component model and a residual model. The principal component model is the product of the principal score matrix and the transpose of the principal loading matrix, while the residual model is the product of the residual score matrix and the transpose of the residual loading matrix.

[0143] S4.3 Perform singular value decomposition on the covariance matrix of the standardized matrix to obtain multiple eigenvalues. Sort the eigenvalues ​​from largest to smallest and calculate the cumulative percentage of the first p1 eigenvalues. Make the cumulative percentage of the first p1 eigenvalues ​​just greater than the set limit to obtain the corresponding p1. Then construct the principal component loading matrix based on the first p1 eigenvalues ​​and construct the residual loading matrix based on the remaining eigenvalues.

[0144] S4.4 Calculate the projection of the normalized matrix onto the residual loading matrix to obtain the residual score matrix;

[0145] S4.5. The Q statistic of the standardized matrix is ​​obtained by multiplying the transpose of the residual score matrix, the residual loading matrix, and the transpose of the standardized matrix; if Q < Q c If Q > Q, then it is determined that the three-phase current transformer is operating normally at this time. c If so, it is determined that the error state of the three-phase current transformer is in an abnormal state.

[0146] Specifically, the online error detection system for current transformers based on secondary data correction also includes a training module for model training. The training module is used for:

[0147] After testing b test data points in the current round, calculate the Q statistic for each of the b test data points in the current round that exceeds the statistic control limit. c If m < 30%b, discard b test data points from the first round of current data, update the next round of current data test data, and continue executing S2; if m ≥ 30%b, discard b test data points from the current round, update the next round of current data test data b test data, and continue executing the first correction module.

[0148] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for online detection of current transformer errors based on secondary data correction, characterized in that, include: S1. Collect historical data and real-time test data from the current transformer to form current data; S2. Calculate the three-phase unbalance sequence based on the current data, calculate the current standard deviation using the three-phase unbalance sequence and compare it with the data correction standard. If the current standard deviation is greater than the data correction standard, then the current data is corrected once to obtain the corrected data. S2.

1. The three-phase unbalance sequence is obtained by taking the absolute value of the difference between the A, B, and C phase data of the current data and the average three-phase current data, and dividing the maximum value of the difference by the average three-phase current data. ; S2.2, Utilizing the three-phase unbalance sequence and the three-phase imbalance sequence CUF The standard deviation of the current is obtained by calculating the standard deviation of the average value. S2.3 Determine the data correction standards; S2.

4. Compare the current standard deviation with the data correction standard. If the current standard deviation is less than or equal to the data correction standard, no correction is needed for the current data. If the current standard deviation is greater than the data correction standard, the current data is corrected once to obtain corrected data. S2.4 includes: S2.4.1 If the standard deviation of the current is less than the data correction standard, it is determined that the current data does not need to be corrected, and S2.4.3 is executed; if the standard deviation of the current is greater than the data correction standard, it is determined that the current data needs to be corrected, and S2.4.2 is executed. S2.4.2 Perform an affine transformation on the current data, use the result of the affine transformation as the current data, and return to S2.1; the affine transformation is to multiply the current data by the linear transformation parameter and then add the translation transformation parameter; S2.4.3 Output the current current data as the first correction data of the current data; S3. Calculate the historical current deviation of each phase of the historical data, perform empirical wavelet transform on it, and establish an autoregressive moving average prediction model to predict the theoretical predicted value of the historical current deviation of each phase; invert the theoretical predicted value of the historical current deviation of each phase to obtain the theoretical predicted value of the historical current data; perform a second correction on the first correction data of the current data based on the theoretical predicted value of the historical current data to obtain the second correction data. S3.1 Subtract the average three-phase current of the historical data from the historical data for each phase data, and then divide by the average three-phase current of the historical data to obtain the historical current deviation of each phase of the historical data; S3.

2. The historical current deviation of each phase in the historical data is obtained by empirical wavelet transform to obtain the modal components of the historical current deviation of each phase. S3.3 Establish an autoregressive moving average prediction model, use the historical current deviation modal components of each phase as the past observations of the autoregressive moving average prediction model, and predict the theoretical values ​​of the historical current deviation of each phase. S3.

4. Perform inversion calculation on the theoretical predicted value of the historical current deviation of each phase. The inversion calculation formula is the theoretical predicted value of the historical current data minus the average value of the three-phase current of the historical data, and then divided by the average value of the three-phase current of the historical data. This equals the theoretical predicted value of the historical current deviation of each phase. Substitute the theoretical predicted value of the historical current deviation of each phase into the inversion calculation formula to obtain the theoretical predicted value of the historical current data. S3.

5. Based on the theoretical prediction values ​​from historical current data, the first-correction data of the current data is further corrected to obtain the second-correction data; S3.5 includes: S3.5.

1. The theoretical predicted value of the historical current data is divided by the test data, and the three-phase average of the quotient is calculated as the mean value characteristic of the quotient. S3.5.

2. Multiply the mean characteristic of the quotient with the first-correction data to obtain the second-correction data of the current data; S4. Calculate using the second-correction data Q Statistics, Q Statistics and Control Limits Q c By comparing the data, it can be determined whether the error state of the current transformer is abnormal.

2. The online error detection method for current transformers based on secondary data correction according to claim 1, characterized in that, S4 includes: S4.

1. Standardize the data after secondary correction to obtain a standardized matrix; S4.2 Decompose the standardized matrix using principal component analysis, resulting in a principal component model and a residual model. The principal component model is the product of the principal score matrix and the transpose of the principal loading matrix, while the residual model is the product of the residual score matrix and the transpose of the residual loading matrix. S4.

3. Perform singular value decomposition on the covariance matrix of the standardized matrix to obtain multiple eigenvalues. Sort the eigenvalues ​​from largest to smallest and calculate the eigenvalues ​​before... The cumulative percentage of the eigenvalues ​​makes the first The cumulative percentage calculated from each feature value is exactly greater than the set limit, thus yielding the corresponding... ， Then based on the previous The principal component loading matrix is ​​constructed from the eigenvalues. ， The remaining eigenvalues ​​are used to construct the residual loading matrix; S4.4 Calculate the projection of the normalized matrix onto the residual loading matrix to obtain the residual score matrix; S4.

5. The product of the transpose of the residual score matrix, the residual loading matrix, and the transpose of the normalized matrix yields the normalized matrix. Q Statistic; if Q < Q c If so, it is determined that the three-phase current transformer is operating normally at this time. Q > Q c If so, it is determined that the error state of the three-phase current transformer is in an abnormal state.

3. The online error detection method for current transformers based on secondary data correction according to claim 2, characterized in that, It also includes S5 for model training, and S5 includes: After the current round is completed b Each test data point is used to statistically analyze the current round. b The Q statistic exceeded the control limit in all test data. Q c Number of times m ,like m <30% b Then discard the first round of current data. b The test data is updated, and the next round of test data for the current data is updated. Then, S2 is executed. m ≥30% b Then discard the current round's... b One test data point, and update the current data for the next round. b Given one test data point, continue executing S2.

4. An online error detection system for current transformers based on secondary data correction, characterized in that, include: The data acquisition module is used to collect historical data and real-time test data from the current transformer to form current data. The first correction module is used to calculate the three-phase unbalance sequence based on current data, calculate the current standard deviation using the three-phase unbalance sequence and compare it with the data correction standard. If the current standard deviation is greater than the data correction standard, the current data is corrected once to obtain corrected data. The first correction module is also used for: S2.

1. The three-phase unbalance sequence is obtained by taking the absolute value of the difference between the A, B, and C phase data of the current data and the average three-phase current data, and dividing the maximum value of the difference by the average three-phase current data. ; S2.2, Utilizing the three-phase unbalance sequence and the three-phase imbalance sequence CUF The standard deviation of the current is obtained by calculating the standard deviation of the average value. S2.3 Determine the data correction standards; S2.

4. Compare the current standard deviation with the data correction standard. If the current standard deviation is less than or equal to the data correction standard, no correction is needed for the current data. If the current standard deviation is greater than the data correction standard, the current data is corrected once to obtain corrected data. S2.4 includes: S2.4.1 If the standard deviation of the current is less than the data correction standard, it is determined that the current data does not need to be corrected, and S2.4.3 is executed; if the standard deviation of the current is greater than the data correction standard, it is determined that the current data needs to be corrected, and S2.4.2 is executed. S2.4.2 Perform an affine transformation on the current data, use the result of the affine transformation as the current data, and return to S2.1; the affine transformation is to multiply the current data by the linear transformation parameter and then add the translation transformation parameter; S2.4.3 Output the current current data as the first correction data of the current data; The second correction module is used to calculate the historical current deviation of each phase in the historical data, perform empirical wavelet transform on it, and establish an autoregressive moving average prediction model to predict the theoretical predicted value of the historical current deviation of each phase; inversely calculate the theoretical predicted value of the historical current data by retrieving the theoretical predicted value of the historical current deviation of each phase; and perform a second correction on the first-correction data of the current data based on the theoretical predicted value of the historical current data to obtain the second-correction data; the second correction module is also used for: S3.1 Subtract the average three-phase current of the historical data from the historical data for each phase data, and then divide by the average three-phase current of the historical data to obtain the historical current deviation of each phase of the historical data; S3.

2. The historical current deviation of each phase in the historical data is obtained by empirical wavelet transform to obtain the modal components of the historical current deviation of each phase. S3.3 Establish an autoregressive moving average prediction model, use the historical current deviation modal components of each phase as the past observations of the autoregressive moving average prediction model, and predict the theoretical values ​​of the historical current deviation of each phase. S3.

4. Perform inversion calculation on the theoretical predicted value of the historical current deviation of each phase. The inversion calculation formula is the theoretical predicted value of the historical current data minus the average value of the three-phase current of the historical data, and then divided by the average value of the three-phase current of the historical data. This equals the theoretical predicted value of the historical current deviation of each phase. Substitute the theoretical predicted value of the historical current deviation of each phase into the inversion calculation formula to obtain the theoretical predicted value of the historical current data. S3.

5. Based on the theoretical prediction values ​​from historical current data, the first-correction data of the current data is further corrected to obtain the second-correction data; S3.5 includes: S3.5.

1. The theoretical predicted value of the historical current data is divided by the test data, and the three-phase average of the quotient is calculated as the mean value characteristic of the quotient. S3.5.

2. Multiply the mean characteristic of the quotient with the first-correction data to obtain the second-correction data of the current data; The online detection module is used to calculate using secondary correction data. Q Statistics, Q Statistics and Control Limits Q c By comparing the data, it can be determined whether the error state of the current transformer is abnormal.

Images