Non-stop intelligent diagnosis method and system for capacitor voltage transformer

Abstract

The present invention provides a non-stop intelligent diagnosis method and system for a capacitive voltage transformer, the method comprising: synchronously collecting t The secondary output signal of the three-phase capacitive voltage transformer at the same measuring point at the same time is used to construct the voltage amplitude data matrix; based on the voltage amplitude data matrix, the voltage amplitude data matrix is ​​detected by canonical variable analysis. t The measurement error state at the moment can realize the self-detection of the overall measurement error of the three-phase capacitor voltage transformer; if t When the overall measurement error at the moment is greater than the preset error threshold, the coding-classification method is used to realize the self-diagnosis of the operating status of the capacitor voltage transformer through correlation analysis and data mining algorithm; the virtual instrument LabVIEW is used to visualize the output t The present invention solves the technical problems of difficulty in real-time monitoring of the operating state of a three-phase capacitor voltage transformer and low efficiency and accuracy in diagnosing fault forms and fault points while detecting the overall error of the three phases.

Description

Technical Field

[0001] The present invention relates to the field of power equipment status assessment, and in particular to a non-stop intelligent diagnosis method and system for a capacitive voltage transformer. Background Art

[0002] Capacitor voltage transformers are widely used measurement equipment in high-voltage power grids, providing accurate primary voltage signals to secondary metering, control and protection equipment. The long-term stability of metering errors is a key characteristic quantity for measuring the operating status of CVT (Capacitor Voltage Transformer). Compared with traditional electromagnetic voltage transformers, the structure of CVT is more complex. Under the influence of the coupling of factors such as grid frequency, natural environment, complex electric field environment and secondary load, the capacitance voltage division characteristics of CVT are prone to randomness, multiple occurrences and diversity in metering errors, resulting in errors in the measurement of primary voltage signals. In order to avoid economic losses and safety hazards in the operation of power systems, it is necessary to effectively detect the abnormal changes in the CVT metering error.

[0003] With the continuous construction of smart grids, more and more power equipment status assessments are developing in the direction of intelligence and systematization. In the field of high-voltage transformer assessment, several representative online assessment methods have emerged. For example, the Chinese invention patent application document "A method and system for evaluating the comprehensive error state of a capacitive voltage transformer" with publication number CN116840764A, the existing method uses a signal processing-based method to evaluate the error state of a capacitive voltage transformer online, but the premise of such a method is to use a stable measured signal as a comparison standard, and the primary physical signal of the power system is a non-stationary random signal with a fluctuation range far greater than 0.05%, so it is difficult to accurately evaluate the transformer operation error fluctuation with a variation range of 0.2%; and the Chinese invention patent application document "A method and device for evaluating the error state of an electronic voltage transformer" with publication number CN115951292A, the existing method can achieve high-precision detection in a few specific scenarios, but the detection standard is not clear, and the above invention does not fully mine the measurement data, and fails to make full use of the data features to further diagnose the operating state of the high-voltage transformer. In the routine maintenance of substations, when abnormal conditions are found in high-voltage equipment, clarifying the fault point and fault type helps to quickly cut off hidden dangers and guide further replacement and repair. For example, the Chinese invention patent application document "A method and system for evaluating the error state of a capacitive voltage transformer" with publication number CN109444791A uses a contribution graph-based method to determine the fault phase, but the contribution graph may cause misjudgment of the fault point due to the fault smearing effect between the variables.

[0004] In summary, the prior art has technical problems such as difficulty in real-time monitoring of the operating status of the three-phase capacitive voltage transformer, and low efficiency and accuracy in diagnosing the fault form and fault point while detecting the overall error of the three phases. Summary of the invention

[0005] The technical problem to be solved by the present invention is: how to solve the technical problem in the prior art that it is difficult to monitor the operating status of a three-phase capacitive voltage transformer in real time, and that the efficiency and accuracy of diagnosing the fault form and fault point while detecting the overall error of the three phases are low.

[0006] The present invention adopts the following technical solution to solve the above technical problem: the non-stop intelligent diagnosis method of the capacitive voltage transformer comprises:

[0007] S1. t At the same measuring point at the same time, the secondary output signal of the three-phase capacitive voltage transformer is synchronously collected, and the voltage amplitude data matrix is ​​constructed based on the secondary output signal X ;

[0008] S2, based on voltage amplitude data matrix X , for the secondary output signal of the three-phase capacitor voltage transformer under normal operation, the three-phase phase correlation is obtained as a virtual standard for overall error detection, and self-detection is performed through standard variable analysis. t The measurement error state at the moment is used to obtain the overall measurement error of the three-phase capacitor voltage transformer and the overall operation state self-detection result;

[0009] S3, in t When the overall measurement error at a certain moment is greater than the preset error threshold, the measurement error characteristic parameters are extracted through correlation analysis, and the structural characteristics and distribution characteristics of the measurement error characteristic parameters are mined by a data mining algorithm to obtain the distribution characteristics of the normal state error characteristic parameters as a virtual comparison standard for fault diagnosis. The coding-classification method is used to perform a self-diagnosis operation on the operating state of the three-phase capacitor voltage transformer to obtain the fault state diagnosis result.

[0010] S4, use preset virtual devices, visual output t The overall operating status self-detection results and fault status diagnosis results of the three-phase capacitor voltage transformer at the moment.

[0011] The present invention can perform continuous electrical intelligent diagnosis operations on three-phase capacitor voltage transformers under the rigid constraints of the three-phase four-wire wiring mode of the power system, monitor the operating status of the three-phase capacitor voltage transformer in real time, and detect the overall measurement error of the three-phase capacitor voltage transformer while being able to quickly and accurately diagnose the fault form and fault point, which helps to promote the further development of power equipment status assessment in the direction of intelligence. The present invention overcomes the limitations of standard transformers and instruments in the traditional detection process, uses fault diagnosis virtual comparison standards for virtual comparison, and uses canonical variable analysis, data mining algorithms, and coding-classification methods. Only one data acquisition device and one data processing terminal are required to realize intelligent diagnosis of three-phase capacitor voltage transformers.

[0012] In a more specific technical solution, in S1, the voltage amplitude data matrix is ​​expressed using the following logic: :

[0013]

[0014] in, n Indicates the number of sampling points of the secondary output voltage of the three-phase capacitor voltage transformer, Indicates the voltage amplitude data of the A phase output of the three-phase capacitor voltage transformer. Indicates the voltage amplitude data of the B phase output of the three-phase capacitor voltage transformer. Indicates the voltage amplitude data of the C phase output of the three-phase capacitor voltage transformer. i Represents the voltage amplitude data matrix The number of rows of elements, where .

[0015] The present invention adopts a virtual standard and virtual instrument technology without power outages, and only detects the overall measurement error online through the secondary output data of the three-phase capacitor voltage transformer. When the measurement error state of the three-phase capacitor voltage transformer is abnormal, the fault point and fault form are quickly diagnosed, and the intelligent diagnosis of the operating state of the three-phase capacitor voltage transformer without power outages is realized. The present invention helps to timely grasp the real-time state of the output data of the three-phase capacitor voltage transformer, effectively reduces the work tasks of the front-line maintenance personnel, and improves the economic benefits of the maintenance work.

[0016] In a more specific technical solution, S2 includes:

[0017] S21. Based on canonical variable analysis, an overall error self-detection model is established for the three-phase capacitive voltage transformer, in which the voltage amplitude data matrix is ​​decomposed and processed using the following logic: X :

[0018]

[0019] In the formula, Z is the canonical variate matrix, E is the residual matrix, is the past matrix, p represents the length of past observations, J is the transformation matrix of the canonical variable space, L is the transformation matrix of the residual space, is the transformation matrix of the past residual space;

[0020] S22, using the historical data of the normal operating state to calculate the overall error detection threshold of the three-phase capacitive voltage transformer , T represents statistical parameters, Indicates confidence, Indicates confidence Corresponding statistical parameters;

[0021] S23, synchronously collect the three-phase output online data, use the overall error self-detection model to self-detect the three-phase capacitive voltage transformer to obtain the overall measurement error, wherein the new past matrix is ​​generated by the Hankel matrix according to the three-phase output online data , use the following logic to transform the matrix and get the new standard variable matrix , the new residual matrix :

[0022]

[0023]

[0024] The following logic is used to calculate the statistics of the three-phase output online data: :

[0025]

[0026] In the formula, r The number of dimensions reserved for the canonical variate space, for k The variables in the moment normalization matrix, i Represents the voltage amplitude data matrix The number of rows of elements.

[0027] In a more specific technical solution, S21 also includes:

[0028] At the moment k , using the following logic to define the past measurement matrix and future measurement matrix :

[0029]

[0030]

[0031] In the formula, p represents the length of past observations, f represents the length of future observations, Represents the past measurement matrix The first measured component of Represents the past measurement matrix The second measurement component of Represents the past measurement matrix No. p Measuring weight, Represents the future measurement matrix The first measured component of Represents the future measurement matrix The second measurement component of Represents the future measurement matrix No. f-1 Measuring weight;

[0032] Using the following formula, the past measurement matrix , Future Measurement Matrix To standardize:

[0033]

[0034]

[0035] In the formula, Represents the past measurement matrix The augmented matrix of Represents the future measurement matrix The augmented matrix of Represents the past measurement matrix The standardized matrix, Represents the future measurement matrix The standardized matrix of

[0036] All past and future vectors in the standardized matrix are summarized and arranged in different columns, and the past matrix is ​​regenerated through the Hankel matrix 、Future Matrix :

[0037]

[0038]

[0039] In the formula, Represents the past measurement matrix No. p+1A standardized matrix, Represents the past measurement matrix No. p+2 A standardized matrix, Represents the past measurement matrix No. p+M A standardized matrix, M is the number of columns of the Hankel matrix, Represents the future measurement matrix No. p+1 A standardized matrix, Represents the future measurement matrix No. p+2 A standardized matrix, Represents the future measurement matrix No. p+M A standardized matrix;

[0040] Using the following logic, calculate the past matrix 、Future Matrix The covariance matrix and cross-covariance matrix of are:

[0041]

[0042]

[0043]

[0044] In the formula, is the number of columns of the Hankel matrix, N is the number of observations for each variable, Represents the past matrix The transposed matrix of Represents the future matrix The transposed matrix of Represents the past matrix The processed covariance matrix is Represents the future matrix The processed covariance matrix is Represents the past matrix 、Future Matrix The processed cross-covariance matrix;

[0045] Using the following logic, the scaled Hankel matrix H The singular value decomposition of maximizes the correlation between past and future observations:

[0046]

[0047] In the formula, U is the first singular vector orthogonal matrix,V is the second singular vector orthogonal matrix, is a diagonal matrix, the first singular vector is an orthogonal matrix U and the second singular vector orthogonal matrix V The correlation between them is given by the diagonal matrix The diagonal elements in express, Represents the future matrix The processed covariance matrix The inverse matrix of the inverse square root of Represents the past matrix The processed covariance matrix The inverse matrix of the inverse square root of Represents the second singular vector orthogonal matrix V The transpose of

[0048] The diagonal elements As singular values, the orthogonal matrix for the singular values ​​and the first singular vector U and the second singular vector orthogonal matrix V Sort the corresponding singular vectors in descending order, select the largest singular value, and obtain the second singular vector orthogonal matrix V The dimension reduction matrix in , r represents the dimensionality reduction operation to obtain the canonical variable space and residual space;

[0049] The transformation matrix that projects past measurements into the canonical variable space uses the following logic: J , the transformation matrix of the residual space L :

[0050]

[0051]

[0052] In the formula, Represents the dimension reduction matrix The transpose of I represents the identity matrix, and uses the following logic to express the canonical variable matrix Z , residual matrix E :

[0053]

[0054] .

[0055] In a more specific technical solution, S22 also includes:

[0056] Under normal operating conditions, the amplitude data of the three-phase capacitor voltage transformer is obtained, and the probability density distribution function is calculated based on the amplitude data using the following kernel density estimation method: f(x) :

[0057]

[0058] In the formula, x is the variable to be estimated, The time under normal working conditions k The sampling point variables, represents the kernel function, K represents the kernel, represents the sum of kernel functions, h is the smoothing factor, M is the number of columns of the Hankel matrix;

[0059] Using the following logic, when the confidence level is In the case of The value of the quantile is statistically the overall error detection threshold :

[0060]

[0061]

[0062] In the formula, Represents the probability density distribution function f(x) The estimation function of Indicates the estimation function (-∞,x] The integral of represents the integral function, Indicates confidence, Represents the inverse function of the integral function in the variable to be estimated x Pick The value at time.

[0063] The present invention designs and clarifies the working principle of the overall error detection virtual standard. On the basis of self-detection of the overall error of the three-phase capacitor voltage transformer, it further quickly and accurately diagnoses the specific error form of the three-phase capacitor voltage transformer under abnormal conditions, thereby realizing non-stop intelligent diagnosis of the three-phase capacitor voltage transformer.

[0064] In more specific technical solutions, S3 includes:

[0065] S31, based on the voltage amplitude data, extracting measurement error characteristic parameters by using principal component analysis in correlation analysis;

[0066] S32, based on the coding-classification method, according to the measurement error characteristic parameters, the autoencoder is used to mine the residual matrixE The characteristic information of the residual space under normal operation is used as the virtual comparison standard for fault diagnosis, and the reconstruction error result is encoded to obtain a multi-dimensional encoding result.

[0067] S33, synchronously collect three-phase output online data, use the error detection model to perform error self-detection on the three-phase capacitive voltage transformer, and complete the operation status self-diagnosis operation;

[0068] S34, based on the coding-classification method, using the learning vector quantization neural network as a classifier, taking the multidimensional coding result of the residual space as input, and each error state as output, to obtain a mapping relationship; inputting the error state into the learning vector quantization neural network to train and obtain a classification model, and adjusting the weight of the winning neuron;

[0069] S35. During the training of the classification model, the initial learning rate Set to a monotonic decreasing function to update the learning rate of the classification model using the following logic :

[0070]

[0071] In the formula, Indicates the number of model training times; represents the maximum number of training times, m Indicates maximum.

[0072] In a more specific technical solution, in S31, the voltage amplitude data is standardized to obtain a voltage amplitude data matrix , for the voltage amplitude data matrix The covariance matrix of R Perform singular value decomposition calculation to obtain the characteristic matrix ;

[0073] The voltage amplitude data matrix is ​​decomposed using the following logic: :

[0074]

[0075] In the formula, A sequence matrix representing voltage amplitude data, represents the transpose operation of the sequence matrix of voltage amplitude data, represents the transpose operation of the feature matrix, e represents the characteristic parameter of measurement error, Represents the sequence matrix of the voltage amplitude data of the first dimension, Indicates n dimensional sequence matrix of voltage amplitude data, is the real value information of the primary voltage, is the residual matrix.

[0076] In a more specific technical solution, S32 includes:

[0077] The residual matrix under normal operation E The characteristic parameters of the measurement errors in each dimension are The relative stability between them is used as the virtual comparison standard for fault diagnosis, and the autoencoder is used to mine the data features of the normal state residual space, where e represents the characteristic parameter of measurement error, Represents the voltage amplitude data matrix The measurement error characteristic parameters of the first row of elements, Represents the voltage amplitude data matrix The measurement error characteristic parameters of the second row elements, Represents the voltage amplitude data matrix The measurement error characteristic parameters of the third row elements;

[0078] Using the normal state residual space data features, the autoencoder is trained to obtain the voltage amplitude data matrix No. i The encoding corresponding to the row element :

[0079]

[0080] In the formula, Represents the voltage amplitude data matrix No. i The measurement error characteristic parameters corresponding to the row elements, w is the weight matrix, b is the bias vector, h represents the transfer function of the autoencoder, the superscript (1) represents the first layer, is the first layer transfer function of the autoencoder, is the first layer weight matrix, is the first layer bias vector;

[0081] Using the self-decoder, the encoded representation Mapping back to the estimated value of the original input vector;

[0082] The decoding operation is performed using the following logic:

[0083]

[0084] In the formula, the superscript (2) indicates the second layer; is the second layer transfer function of the self-decoder, is the second layer weight matrix, is the bias vector of the second layer, Represents the reconstructed value of the characteristic parameter of the measurement error;

[0085] Use the following logic to express the optimization objective function:

[0086]

[0087] In the formula, E is the residual matrix, Reconstruct the matrix of the original data learned by the autoencoder, is the distance metric function between the residual matrix and the original data reconstruction matrix learned by the autoencoder. MinimizeLoss represents minimizing loss;

[0088] The following logic is used to encode the test results of the operating status self-diagnosis operation:

[0089]

[0090] In the formula, code Represents a verification code.

[0091] When the overall error state of the three-phase capacitor voltage transformer is abnormal, the present invention performs correlation analysis and deep data mining on the measurement data output by multiple three-phase capacitor voltage transformers, uses the measurement error characteristic parameters extracted by principal component analysis under normal operation as a virtual comparison standard for fault diagnosis, and uses an autoencoder and a learning vector quantization neural network to realize self-diagnosis of the fault state of the three-phase capacitor voltage transformer, helps to quickly locate the fault point and determine the error change trend, which is beneficial for the staff to accurately grasp the operating state of the three-phase capacitor voltage transformer.

[0092] In a more specific technical solution, in step S34, when the network classification result of the classification model is consistent with the multi-dimensional encoding result, the weight is adjusted in the direction of the input sample using the following logic:

[0093]

[0094] In the formula, express k The time corresponding to t+1 The weight of the winning neuron in the training, express k The time corresponding to t The weight of the winning neuron in the training, represents the winning neuron, represents the learning rate of the classification model, Represents the voltage amplitude data matrix X No.i The variable to be estimated corresponding to the row element;

[0095] When the network classification result is inconsistent with the multi-dimensional encoding result, the following logic is used to adjust the weights in the opposite direction of the input sample:

[0096]

[0097] In a more specific technical solution, the non-stop intelligent diagnosis system for capacitor voltage transformer includes:

[0098] Data acquisition module, used to t At the same measuring point at the same time, the secondary output signal of the three-phase capacitive voltage transformer is synchronously collected, and the voltage amplitude data matrix is ​​constructed based on the secondary output signal ;

[0099] Overall error self-detection module, used to detect the voltage amplitude data matrix , for the secondary output signal of the three-phase capacitor voltage transformer under normal operation, the three-phase phase correlation is obtained as a virtual standard for overall error detection, and self-detection is performed through standard variable analysis. t The measurement error state at the moment is used to obtain the overall measurement error of the three-phase capacitive voltage transformer, and the overall operation state self-detection result is obtained. The overall error self-detection module is connected to the data acquisition module;

[0100] Abnormal state self-diagnosis module, used to t When the overall measurement error at a certain moment is greater than a preset error threshold, the measurement error characteristic parameters are extracted through correlation analysis, and the structural characteristics and distribution characteristics of the measurement error characteristic parameters are mined by a data mining algorithm to obtain the distribution characteristics of the normal state error characteristic parameters as a virtual comparison standard for fault diagnosis. The three-phase capacitive voltage transformer is subjected to a running state self-diagnosis operation using a coding-classification method to obtain a fault state diagnosis result, and the abnormal state self-diagnosis module is connected to the data acquisition module;

[0101] The visualization output module is used to use a preset virtual device to visualize the output of the overall operation status self-detection result and the fault status diagnosis result of the three-phase capacitive voltage transformer at time t. The visualization output module is connected to the abnormal status self-diagnosis module and the overall error self-detection module.

[0102] Compared with the prior art, the present invention has the following advantages:

[0103] The present invention can perform continuous electrical intelligent diagnosis operations on three-phase capacitor voltage transformers under the rigid constraints of the three-phase four-wire wiring mode of the power system, monitor the operating status of the three-phase capacitor voltage transformer in real time, and detect the overall measurement error of the three-phase capacitor voltage transformer while being able to quickly and accurately diagnose the fault form and fault point, which helps to promote the further development of power equipment status assessment in the direction of intelligence. The present invention overcomes the limitations of standard transformers and instruments in the traditional detection process, uses fault diagnosis virtual comparison standards for virtual comparison, and uses canonical variable analysis, data mining algorithms, and coding-classification methods. Only one data acquisition device and one data processing terminal are required to realize intelligent diagnosis of three-phase capacitor voltage transformers.

[0104] The present invention adopts a virtual standard and virtual instrument technology without power outages, and only detects the overall measurement error online through the secondary output data of the three-phase capacitor voltage transformer. When the measurement error state of the three-phase capacitor voltage transformer is abnormal, the fault point and fault form are quickly diagnosed, and the intelligent diagnosis of the operating state of the three-phase capacitor voltage transformer without power outages is realized. The present invention helps to timely grasp the real-time state of the output data of the three-phase capacitor voltage transformer, effectively reduces the work tasks of the front-line maintenance personnel, and improves the economic benefits of the maintenance work.

[0105] The present invention designs and clarifies the working principle of the overall error detection virtual standard. On the basis of self-detection of the overall error of the three-phase capacitor voltage transformer, it further quickly and accurately diagnoses the specific error form of the three-phase capacitor voltage transformer under abnormal conditions, thereby realizing non-stop intelligent diagnosis of the three-phase capacitor voltage transformer.

[0106] When the overall error state of the three-phase capacitor voltage transformer is abnormal, the present invention performs correlation analysis and deep data mining on the measurement data output by multiple three-phase capacitor voltage transformers, uses the measurement error characteristic parameters extracted by principal component analysis under normal operation as a virtual comparison standard for fault diagnosis, and uses an autoencoder and a learning vector quantization neural network to realize self-diagnosis of the fault state of the three-phase capacitor voltage transformer, helps to quickly locate the fault point and determine the error change trend, which is beneficial for the staff to accurately grasp the operating state of the three-phase capacitor voltage transformer.

[0107] The present invention solves the technical problems existing in the prior art that it is difficult to monitor the operating status of a three-phase capacitive voltage transformer in real time, and that the efficiency and accuracy of diagnosing fault forms and fault points while detecting the three-phase overall error are low. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] Figure 1 A schematic diagram of basic steps of a non-stop intelligent diagnosis method for a capacitor voltage transformer according to Embodiment 1 of the present invention;

[0109] Figure 2 This is a schematic diagram of specific steps for performing overall measurement error self-detection of a three-phase capacitor voltage transformer according to Embodiment 1 of the present invention;

[0110] Figure 3 This is a schematic diagram of data flow processing for overall measurement error self-detection of a three-phase capacitor voltage transformer according to Embodiment 1 of the present invention;

[0111] Figure 4 This is a schematic diagram of specific steps for implementing self-diagnosis of the operating status of a capacitor voltage transformer using a coding-classification method according to Embodiment 1 of the present invention;

[0112] Figure 5 The visual output of Example 1 of the present invention t A schematic diagram of the specific steps of the real-time operating status of the capacitor voltage transformer at each moment;

[0113] Figure 6 This is a LabVIEW-based visualization interface and structural diagram of intelligent diagnosis of a capacitive voltage transformer according to Example 1 of the present invention. DETAILED DESCRIPTION

[0114] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0115] Example 1

[0116] like Figure 1 As shown, the non-stop intelligent diagnosis method for capacitor voltage transformer provided by the present invention comprises the following basic steps:

[0117] Step S1: t At the same measuring point at the same time, the secondary output signal of the three-phase capacitive voltage transformer is synchronously collected, and the voltage amplitude data matrix is ​​constructed based on the secondary output signal ;

[0118] In this embodiment, the voltage amplitude data matrix is ​​constructed as described above. In step S1, the standard amplitude data matrix collected synchronously It is expressed as the following formula:

[0119]

[0120] in, nIndicates the number of sampling points of the secondary output voltage of the three-phase capacitor voltage transformer, Indicates the voltage amplitude data of the A phase output of the three-phase capacitor voltage transformer. Indicates the voltage amplitude data of the B phase output of the three-phase capacitor voltage transformer. Indicates the voltage amplitude data of the C phase output of the three-phase capacitor voltage transformer. i Represents the voltage amplitude data matrix The number of rows of elements, where .

[0121] Step S2: Based on the voltage amplitude data matrix , for the secondary output signal of the three-phase capacitor voltage transformer under normal operation, the three-phase phase correlation is obtained as a virtual standard for overall error detection, and self-detection is performed through standard variable analysis. t The measurement error state at the moment is used to obtain the overall measurement error of the three-phase capacitor voltage transformer and the overall operation state self-detection result;

[0122] like Figure 2 and Figure 3 As shown, in this embodiment, the step S2 of self-detecting the overall measurement error of the three-phase capacitor voltage transformer also includes the following specific steps:

[0123] S21. Based on canonical variable analysis, an overall error self-detection model is established for the three-phase capacitor voltage transformer;

[0124] In this embodiment, data of a three-phase capacitor voltage transformer in a healthy operating state is collected, and an overall error self-detection model for the three-phase capacitor voltage transformer is established by using a method of canonical variable analysis;

[0125] In this embodiment, the voltage amplitude data matrix of the three-phase capacitor voltage transformer in a healthy operating state is collected. X , the voltage amplitude data matrix X Decomposed into the following formula:

[0126]

[0127] Specifically, Z is the canonical variable matrix. The information represented by the canonical variable matrix includes but is not limited to: measurement error information, maximization of original information, is the residual matrix. The information represented by the residual matrix includes but is not limited to: capacitance voltage transformer measurement noise, Yp is the past matrix, which is generated using the original measurement data;

[0128] In this embodiment, Jis the transformation matrix of the canonical variable space. By projecting the past measurement values ​​into the canonical variable space, we get the transformation matrix of the canonical variable space. J , the specific calculation method of the above transformation matrix includes:

[0129] Take any time k For the current moment, define the past measurement matrix and future measurement matrix for:

[0130]

[0131]

[0132] in, p represents the length of past observations, f Represents the length of future observations, which satisfies the following conditions under normal working conditions: p = f .

[0133] Using the following formula, the past measurement matrix , Future Measurement Matrix To standardize:

[0134]

[0135]

[0136] All past and future vectors in the standardized matrix are summarized and arranged in different columns, and the past matrix is ​​regenerated through the Hankel matrix 、Future Matrix :

[0137]

[0138]

[0139] In this embodiment, is the number of columns of the Hankel matrix, N is the number of observations for each variable.

[0140] In this embodiment, the past matrix is ​​calculated using the following formula: 、Future Matrix The covariance matrix and cross-covariance matrix of :

[0141]

[0142]

[0143]

[0144] In this embodiment, by scaling the Hankel matrix The singular value decomposition of maximizes the correlation between past and future observations:

[0145]

[0146] in, U is the first singular vector orthogonal matrix, V is the second singular vector orthogonal matrix, the first singular vector orthogonal matrix U and the second singular vector orthogonal matrix V Pairwise correlation. The first singular vector orthogonal matrix U and the second singular vector orthogonal matrix V The correlation between them is given by the diagonal matrix The diagonal elements in Represents. The diagonal elements As singular values, the orthogonal matrix for the singular values ​​and the first singular vector U and the second singular vector orthogonal matrix V Sort the corresponding singular vectors in descending order, select the largest singular value, and the orthogonal matrix corresponding to the second singular vector V The dimension reduction matrix composed of the column vectors in , obtain the canonical variable space, and take the remaining column vectors to form the residual space.

[0147] In this embodiment, the transformation matrix that projects the past measurements into the canonical variate space is J :

[0148]

[0149] In this embodiment, the transformation matrix that projects the past measurement values ​​to the residual space is L :

[0150]

[0151] In this embodiment, the canonical variable matrix And as follows:

[0152]

[0153] Residual Matrix As follows:

[0154] .

[0155] S22, using the historical data of the normal operating state to calculate the overall error detection threshold of the three-phase capacitive voltage transformer ;

[0156] in, T represents statistical parameters, Indicates confidence, Indicates confidence Corresponding statistical parameters;

[0157] In this embodiment, under normal operating conditions, the amplitude data of the three-phase capacitor voltage transformer is obtained, and the probability density distribution function is calculated based on the amplitude data using the following kernel density estimation method: f(x) ;

[0158] In this embodiment, according to the amplitude data of the three-phase capacitor voltage transformer under normal operating conditions, the probability density distribution function is calculated using the kernel density estimation method. f(x) The expression is as follows:

[0159]

[0160] in, x is the variable to be estimated, The time under normal working conditions k The sampling point variables, represents the kernel function;

[0161] In this embodiment, for example, a Gaussian kernel function may be selected, where: h is the smoothing factor. Statistical overall error detection threshold The calculation formula is as follows:

[0162]

[0163]

[0164] Among them, the above calculation formula means that when the confidence level is In the case of The quantile statistics of is the overall error detection threshold ;

[0165] S23, synchronously collect three-phase output online data, and use the overall error self-detection model to perform self-detection on the three-phase capacitive voltage transformer to obtain the overall measurement error;

[0166] In this embodiment, the three-phase output online data of the three-phase capacitor voltage transformer is collected, and a new past matrix is ​​generated through the Hankel matrix , the new canonical variable matrix obtained by transforming the matrix , the new residual matrix :

[0167]

[0168]

[0169] In this embodiment, the statistics of the three-phase output online data are calculated. :

[0170]

[0171] in, r The number of dimensions reserved for the canonical variate space, for k Moments normalize the variables in the matrix.

[0172] In this embodiment, if the statistic and the overall error detection threshold satisfy: , indicating that the monitoring status of the overall measurement error is normal, otherwise the monitoring status of the overall measurement error is abnormal, and the operating status of the three-phase capacitor voltage transformer is in an abnormal operating state.

[0173] Step S3: t When the overall measurement error at a certain moment is greater than the preset error threshold, the measurement error characteristic parameters are extracted through correlation analysis, and the structural characteristics and distribution characteristics of the measurement error characteristic parameters are mined by a data mining algorithm to obtain the distribution characteristics of the normal state error characteristic parameters as a virtual comparison standard for fault diagnosis. The coding-classification method is used to perform a self-diagnosis operation on the operating state of the three-phase capacitor voltage transformer to obtain the fault state diagnosis result.

[0174] like Figure 4 and Figure 5 As shown, in this embodiment, the step S3 of implementing the self-diagnosis of the operating state of the three-phase capacitor voltage transformer by using the coding-classification method also includes the following specific steps:

[0175] S31, based on the voltage amplitude data, extracting measurement error characteristic parameters by using principal component analysis in correlation analysis;

[0176] In this embodiment, the voltage amplitude data is standardized to obtain a voltage amplitude data matrix , for the voltage amplitude data matrix The covariance matrix of R Perform singular value decomposition calculation to obtain the characteristic matrix ;

[0177] In this embodiment, the voltage amplitude data matrix Decomposed into the following formula:

[0178]

[0179] in, is the real value information of the primary voltage, is the residual matrix. In this embodiment, the residual matrix The method also includes: error characteristic parameters, which are used to characterize the comprehensive error information of the three-phase capacitive voltage transformer at the same measuring point;

[0180] In this embodiment, the residual matrix The specific form can be expressed as follows:

[0181] ;

[0182] S32, based on the coding-classification method, according to the measurement error characteristic parameters, the autoencoder is used to mine the residual matrix E The characteristic information of the residual space under normal operation is used as the virtual comparison standard for fault diagnosis, and the reconstruction error result is encoded to obtain a multi-dimensional encoding result.

[0183] In this embodiment, the residual matrix under normal operation is E The characteristic parameters of the measurement errors in each dimension are The relative stability between them is used as the virtual comparison standard for fault diagnosis, and the autoencoder is used to mine the data characteristics of the residual space of the normal state;

[0184] In this embodiment, the normal state residual space data features are used to train the autoencoder to obtain the voltage amplitude data matrix No. i The encoding corresponding to the row element :

[0185]

[0186] The superscript (1) indicates the first layer. is the first layer transfer function of the autoencoder, is the first layer weight matrix, is the first layer bias vector. Then, using the self-decoder, the encoded representation Mapping back to the estimated value of the original input vector;

[0187] The decoding process of this embodiment is as follows:

[0188]

[0189] Among them, the superscript (2) indicates the second layer; is the second layer transfer function of the self-decoder, is the second layer weight matrix, is the bias vector of the second layer.

[0190] In this embodiment, the optimization objective function is as follows:

[0191]

[0192] in, Reconstruct the matrix of the original data learned by the autoencoder, It is the distance metric function between the residual matrix and the reconstruction matrix of the original data learned by the autoencoder.

[0193] In this embodiment, the detection result of the running state self-diagnosis operation is encoded according to the following formula to improve the efficiency and accuracy of the intelligent diagnosis of the error state of the three-phase capacitor voltage transformer:

[0194]

[0195] S33, synchronously collect three-phase output online data, use the error detection model to perform error self-detection on the three-phase capacitive voltage transformer, and complete the operation status self-diagnosis operation;

[0196] S34, based on the coding-classification method, using the learning vector quantization neural network as a classifier, taking the multidimensional coding result of the residual space as input, and each error state as output, to obtain a mapping relationship; inputting the error state into the learning vector quantization neural network to train and obtain a classification model, and adjusting the weight of the winning neuron;

[0197] In this embodiment, the classification-based diagnosis operation includes: training a classification model;

[0198] In this embodiment, in the operation of training the classification model, the error state is input into the learning vector quantization neural network to train the classification model, and whether the classification is correct is determined. The weight of the winning neuron is adjusted according to the difference rule. When the network classification result is consistent with the multidimensional encoding result, the weight is adjusted in the direction of the input sample:

[0199]

[0200] When the classification result is inconsistent with the multidimensional encoding result, adjust the weights in the opposite direction of the input sample:

[0201]

[0202] in, Represents the voltage amplitude data matrix X No. i The variables to be estimated corresponding to the row elements are taken as multidimensional input vectors, express k The time corresponding to t The weight of the winning neuron in the training is a multi-dimensional prototype vector.

[0203] S35. During the training of the classification model, the initial learning rate Set to a monotonic decreasing function to update the learning rate of the classification model using the following logic :

[0204]

[0205] In the formula, Indicates the number of model training times; represents the maximum number of training times, m Indicates maximum.

[0206] In this embodiment, a learning vector quantization neural network is used as a classifier, the multidimensional encoding result of the residual space is used as input, and each error state is used as output, and the mapping relationship is:

[0207]

[0208] in, code Represents the verification code, n Represents the number of sampling points of the secondary output voltage of the three-phase capacitor voltage transformer.

[0209] In this embodiment, the diagnosis content of the three-phase capacitor voltage transformer includes but is not limited to: high risk error, A phase output is too large, B phase output is too large, C phase output is too large, A phase output is too small, B phase output is too small and C phase output is too small. In this embodiment, the weights of the remaining non-winning neurons remain unchanged, and the learning rate is guaranteed through the aforementioned training process. is a monotonically decreasing function and updates the learning rate.

[0210] When the classification model training is completed, t The encoding result of the moment residual space reconstruction error is input into the trained learning vector quantization neural network to realize the intelligent diagnosis of the specific error state of the three-phase capacitor voltage transformer;

[0211] Step S4: Use the virtual instrument LabVIEW to visualize and output the real-time operating status of the capacitor voltage transformer at time t, including the overall operating status and fault status diagnosis of the three-phase capacitor voltage transformer.

[0212] like Figure 6 As shown, in this embodiment, the visualization output tIn step S4 of the real-time operating state of the three-phase capacitor voltage transformer at the moment, the virtual instrument LabView is used to visualize the self-diagnosis results of the overall error self-detection results of the three-phase capacitor voltage transformer and various error states, and the MATLAB script is loaded into the aforementioned virtual instrument LabVIEW. The received voltage data is analyzed and calculated by the script to obtain the specific error state, and the intelligent diagnosis results of the operating state of the three-phase capacitor voltage transformer are displayed in real time by the front panel of the virtual instrument LabView. In this embodiment, the virtual instrument LabVIEW is used for offline data training, and the training model operation is performed according to different substations and different types of equipment to improve the flexibility of the system. In addition, the visual diagnostic display interface can display the operating state of the three-phase capacitor voltage transformer in real time, realize the intelligent and systematic evaluation of the power equipment state, and simplify the abnormal detection operation method.

[0213] In summary, the present invention can perform continuous electrical intelligent diagnosis operations on three-phase capacitor voltage transformers under the rigid constraints of the three-phase four-wire wiring mode of the power system, monitor the operating status of the three-phase capacitor voltage transformer in real time, and detect the overall measurement error of the three-phase capacitor voltage transformer. At the same time, it can quickly and accurately diagnose the fault form and fault point, which is helpful to promote the further development of power equipment status assessment in the direction of intelligence. The present invention overcomes the limitations of standard transformers and instruments in the traditional detection process, adopts virtual comparison standards for fault diagnosis, and adopts canonical variable analysis, data mining algorithms, and coding-classification methods. Only one data acquisition device and one data processing terminal are required to realize intelligent diagnosis of three-phase capacitor voltage transformers.

[0214] The present invention adopts a virtual standard and virtual instrument technology without power outages, and only detects the overall measurement error online through the secondary output data of the three-phase capacitor voltage transformer. When the measurement error state of the three-phase capacitor voltage transformer is abnormal, the fault point and fault form are quickly diagnosed, and the intelligent diagnosis of the operating state of the three-phase capacitor voltage transformer without power outages is realized. The present invention helps to timely grasp the real-time state of the output data of the three-phase capacitor voltage transformer, effectively reduces the work tasks of the front-line maintenance personnel, and improves the economic benefits of the maintenance work.

[0215] The present invention designs and clarifies the working principle of the overall error detection virtual standard. On the basis of self-detection of the overall error of the three-phase capacitor voltage transformer, it further quickly and accurately diagnoses the specific error form of the three-phase capacitor voltage transformer under abnormal conditions, thereby realizing non-stop intelligent diagnosis of the three-phase capacitor voltage transformer.

[0216] When the overall error state of the three-phase capacitor voltage transformer is abnormal, the present invention performs correlation analysis and deep data mining on the measurement data output by multiple three-phase capacitor voltage transformers, uses the measurement error characteristic parameters extracted by principal component analysis under normal operation as a virtual comparison standard for fault diagnosis, and uses an autoencoder and a learning vector quantization neural network to realize self-diagnosis of the fault state of the three-phase capacitor voltage transformer, helps to quickly locate the fault point and determine the error change trend, which is beneficial for the staff to accurately grasp the operating state of the three-phase capacitor voltage transformer.

[0217] The present invention solves the technical problems existing in the prior art that it is difficult to monitor the operating status of a three-phase capacitive voltage transformer in real time, and that the efficiency and accuracy of diagnosing fault forms and fault points while detecting the three-phase overall error are low.

[0218] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A non-stop intelligent diagnosis method for capacitive voltage transformer, characterized in that: The method comprises: S1. t At the same measuring point at the same time, the secondary output signal of the three-phase capacitive voltage transformer is synchronously collected, and the voltage amplitude data matrix is ​​constructed according to the secondary output signal ; S2, based on the voltage amplitude data matrix , for the secondary output signal of the three-phase capacitive voltage transformer under normal operation, obtain the three-phase phase correlation as the overall error detection virtual standard, and self-detect the t The measurement error state at the moment is used to obtain the overall measurement error of the three-phase capacitor voltage transformer and the overall operation state self-detection result; S3, in the t When the overall measurement error at the time is greater than the preset error threshold, the measurement error characteristic parameters are extracted by correlation analysis, and the structural characteristics and distribution characteristics of the measurement error characteristic parameters are mined by a data mining algorithm to obtain the distribution characteristics of the normal state error characteristic parameters as a virtual comparison standard for fault diagnosis, and the three-phase capacitor voltage transformer is subjected to a running state self-diagnosis operation by using a coding-classification method to obtain a fault state diagnosis result; S4. Use the preset virtual device to visualize the output of the t The overall operating status self-detection result and the fault status diagnosis result of the three-phase capacitive voltage transformer at the time.

2. The non-stop intelligent diagnosis method for capacitive voltage transformer according to claim 1 is characterized in that: In S1, the voltage amplitude data matrix is ​​expressed using the following logic: : In the formula, n represents the number of sampling points of the secondary output voltage of the three-phase capacitive voltage transformer, represents the voltage amplitude data output by the A phase of the three-phase capacitive voltage transformer, represents the voltage amplitude data output by the B phase of the three-phase capacitive voltage transformer, represents the voltage amplitude data output by the C phase of the three-phase capacitive voltage transformer, i The voltage amplitude data matrix is ​​represented by The number of rows of elements, where .

3. The non-stop intelligent diagnosis method for capacitive voltage transformer according to claim 1 is characterized in that: The S2 includes: S21, based on the canonical variable analysis, for the three-phase capacitor voltage transformer, an overall error self-detection model is established, wherein the voltage amplitude data matrix is ​​decomposed and processed using the following logic X : In the formula, Z is the canonical variate matrix, E is the residual matrix, is the past matrix, p represents the length of past observations, J is the transformation matrix of the canonical variable space, L is the transformation matrix of the residual space, is the transformation matrix of the past residual space; S22, calculating the overall error detection threshold of the three-phase capacitor voltage transformer using the historical data of the normal operating state , T represents statistical parameters, Indicates confidence, Indicates confidence The corresponding statistical parameters; S23, synchronously collect the three-phase output online data, use the overall error self-detection model to self-detect the three-phase capacitive voltage transformer to obtain the overall measurement error, wherein, according to the three-phase output online data, a new past matrix is ​​generated through the Hankel matrix , use the following logic to transform the matrix and get the new standard variable matrix , the new residual matrix : The statistics of the three-phase output online data are calculated using the following logic: : In the formula, r The number of dimensions reserved for the canonical variate space, for k The variables in the moment normalization matrix, i The voltage amplitude data matrix is ​​represented by The number of rows of elements.

4. The non-stop intelligent diagnosis method for capacitive voltage transformer according to claim 3 is characterized in that: The S21 further includes: At the moment k , using the following logic to define the past measurement matrix and future measurement matrix : In the formula, p represents the length of past observations, f represents the length of future observations, Represents the past measurement matrix The first measured component of Denotes the past measurement matrix The second measurement component of Represents the past measurement matrix No. p Measuring weight, Denotes the future measurement matrix The first measured component of Denotes the future measurement matrix The second measurement component of Denotes the future measurement matrix No. f-1 Measuring weight; The past measurement matrix is ​​calculated using the following formula: , the future measurement matrix To standardize: In the formula, Represents the past measurement matrix The augmented matrix of Denotes the future measurement matrix The augmented matrix of Represents the past measurement matrix The standardized matrix, Denotes the future measurement matrix The standardized matrix of All past vectors and future vectors in the standardized matrix are summarized and arranged in different columns, and the past matrix is ​​regenerated through the Hankel matrix 、Future Matrix : In the formula, Represents the past measurement matrix No. p+1 A standardized matrix, Represents the past measurement matrix No. p+2 The standardized matrix, Represents the past measurement matrix No. p+M The standardized matrix, M is the number of columns of the Hankel matrix, Denotes the future measurement matrix No. p+1 The standardized matrix, Denotes the future measurement matrix No. p+2 The standardized matrix, Denotes the future measurement matrix No. p+M The standardized matrix; Using the following logic, the past matrix is ​​calculated 、Future Matrix The covariance matrix and cross-covariance matrix of are: In the formula, is the number of columns of the Hankel matrix, N is the number of observations for each variable, Represents the past matrix The transposed matrix of Represents the future matrix The transposed matrix of Represents the past matrix The covariance matrix obtained by processing is Represents the future matrix The covariance matrix obtained by processing is Represents the past matrix 、The future matrix The processed cross-covariance matrix; Using the following logic, the scaled Hankel matrix H The singular value decomposition of maximizes the correlation between past and future observations: In the formula, U is the first singular vector orthogonal matrix, V is the second singular vector orthogonal matrix, is a diagonal matrix, the first singular vector is an orthogonal matrix U and the second singular vector orthogonal matrix V The correlation between The diagonal elements in express, Represents the future matrix The covariance matrix obtained by processing The inverse matrix of the inverse square root of Represents the past matrix The covariance matrix obtained by processing The inverse matrix of the inverse square root of Denote the second singular vector orthogonal matrix V The transpose of The diagonal elements As singular values, the orthogonal matrix of the singular values ​​and the first singular vectors U and the second singular vector orthogonal matrix V Sort the corresponding singular vectors in descending order, select the largest singular value, and obtain the second singular vector orthogonal matrix V The dimension reduction matrix in , r represents a dimensionality reduction operation to obtain the canonical variable space and the residual space; The transformation matrix that projects past measurements to the canonical variable space is: J , the transformation matrix of the residual space L : In the formula, Represents the dimension reduction matrix The transpose of I Represents the identity matrix, and uses the following logic to express the canonical variable matrix Z , the residual matrix E : 。 5. The non-stop intelligent diagnosis method for capacitive voltage transformer according to claim 3 is characterized in that: The S22 further includes: Under the normal operating state, the amplitude data of the three-phase capacitor voltage transformer is obtained, and the probability density distribution function is calculated based on the amplitude data using the following kernel density estimation method: f(x) : In the formula, x is the variable to be estimated, The time under normal working conditions k The sampling point variables, represents the kernel function, K represents the kernel, represents the sum of the kernel function, h is the smoothing factor, M is the number of columns of the Hankel matrix; Using the following logic, when the confidence level is In the case of The value of the quantile is statistically the overall error detection threshold : In the formula, Represents the probability density distribution function f(x) The estimation function of Indicates that the estimation function is (-∞,x] The integral of represents the integral function, Indicates confidence, represents the inverse function of the integral function in the variable to be estimated x Pick The value at time.

6. The non-stop intelligent diagnosis method for capacitive voltage transformer according to claim 1, characterized in that: The S3 includes: S31, based on the voltage amplitude data, extracting the measurement error characteristic parameters by using the principal component analysis in the correlation analysis; S32, based on the encoding-classification method, according to the measurement error characteristic parameters, using an autoencoder to mine the residual matrix E characteristic information, taking the stable structure of the residual space under the normal operating state as the fault diagnosis virtual comparison standard, encoding the reconstruction error result, and obtaining a multi-dimensional encoding result; S33, synchronously collect three-phase output online data, use the error detection model to perform error self-detection on the three-phase capacitive voltage transformer, and complete the operation of self-diagnosis of the operating status; S34, based on the encoding-classification method, using a learning vector quantization neural network as a classifier, taking the multidimensional encoding result of the residual space as input, and each error state as output, to obtain a mapping relationship; inputting the error state into the learning vector quantization neural network to train and obtain a classification model, and adjusting the weight of the winning neuron; S35. During the training process of the classification model, the initial learning rate Set to a monotonic decreasing function to update the learning rate of the classification model using the following logic : In the formula, Indicates the number of model training times; represents the maximum number of training times, m Indicates maximum.

7. The non-stop intelligent diagnosis method for capacitor voltage transformer according to claim 6, characterized in that: In the step S31, the voltage amplitude data is standardized to obtain the voltage amplitude data matrix , for the voltage amplitude data matrix The covariance matrix of R Perform singular value decomposition calculation to obtain the characteristic matrix ; The voltage amplitude data matrix is ​​decomposed using the following logic: : In the formula, represents the sequence matrix of the voltage amplitude data, represents a transpose operation on the sequence matrix of the voltage amplitude data, represents the transpose operation of the feature matrix, e represents the measurement error characteristic parameter, The sequence matrix representing the voltage amplitude data of the first dimension, Indicates n The sequence matrix of the voltage amplitude data of dimension, is the real value information of the primary voltage, is the residual matrix.

8. The non-stop intelligent diagnosis method for capacitor voltage transformer according to claim 6, characterized in that: The S32 includes: The residual matrix under the normal operating state E The measurement error characteristic parameters of each dimension are The relative stability between them is used as the virtual comparison standard for fault diagnosis, and the autoencoder is used to mine the data features of the normal state residual space, wherein, e represents the measurement error characteristic parameter, Represents the voltage amplitude data matrix The first row of elements are the measurement error characteristic parameters, The voltage amplitude data matrix is ​​represented by The measurement error characteristic parameters of the second row elements, The voltage amplitude data matrix is ​​represented by The measurement error characteristic parameters of the third row elements; Using the normal state residual space data features, the autoencoder is trained to obtain the voltage amplitude data matrix No. i The encoding corresponding to the row element : In the formula, The voltage amplitude data matrix is ​​represented by No. i The measurement error characteristic parameters corresponding to the row elements, w is the weight matrix, b is the bias vector, h represents the transfer function of the autoencoder, the superscript (1) represents the first layer, is the first layer transfer function of the autoencoder, is the first layer weight matrix, is the first layer bias vector; Using a self-decoder, the coded representation Mapping back to the estimated value of the original input vector; The decoding operation is performed using the following logic: In the formula, the superscript (2) indicates the second layer; is the second layer transfer function of the self-decoder, is the second layer weight matrix, is the bias vector of the second layer, represents a reconstructed value of the measurement error characteristic parameter; Use the following logic to express the optimization objective function: In the formula, E is the residual matrix, Reconstruct the matrix of the original data learned by the autoencoder, is the distance metric function between the residual matrix and the original data reconstruction matrix learned by the autoencoder, MinimizeLoss represents minimizing loss; The detection result of the operation self-diagnosis operation is encoded using the following logic: In the formula, code Represents a verification code.

9. The non-stop intelligent diagnosis method for capacitive voltage transformer according to claim 6, characterized in that: In S34, when the network classification result of the classification model is consistent with the multi-dimensional encoding result, the weight is adjusted in the direction of the input sample using the following logic: In the formula, express k The time corresponding to t+1 The weight of the winning neuron after training, express k The time corresponding to t The weight of the winning neuron after training, represents the winning neuron, represents the learning rate of the classification model, Represents the voltage amplitude data matrix X No. i The variable to be estimated corresponding to the row element; When the network classification result is inconsistent with the multi-dimensional encoding result, the weight is adjusted in the opposite direction of the input sample using the following logic:

10. Non-stop intelligent diagnosis system for capacitor voltage transformer, characterized in that: The system comprises: Data acquisition module, used to t At the same measuring point at the same time, the secondary output signal of the three-phase capacitive voltage transformer is synchronously collected, and the voltage amplitude data matrix is ​​constructed according to the secondary output signal ; An overall error self-detection module is used to detect the voltage amplitude data matrix based on the voltage amplitude data matrix. , for the secondary output signal of the three-phase capacitive voltage transformer under normal operation, obtain the three-phase phase correlation as the overall error detection virtual standard, and self-detect the t The measurement error state at the moment is used to obtain the overall measurement error of the three-phase capacitor voltage transformer and the overall operation state self-detection result, wherein the overall error self-detection module is connected to the data acquisition module; The abnormal state self-diagnosis module is used to t When the overall measurement error at the time is greater than the preset error threshold, the measurement error characteristic parameters are extracted by correlation analysis, and the structural characteristics and distribution characteristics of the measurement error characteristic parameters are mined by a data mining algorithm to obtain the distribution characteristics of the normal state error characteristic parameters as a virtual comparison standard for fault diagnosis. The three-phase capacitor voltage transformer is subjected to a running state self-diagnosis operation by using a coding-classification method to obtain a fault state diagnosis result, and the abnormal state self-diagnosis module is connected to the data acquisition module; A visualization output module is used to use a preset virtual device to visualize the overall operating status self-detection result and the fault status diagnosis result of the three-phase capacitive voltage transformer at the time t, and the visualization output module is connected to the abnormal status self-diagnosis module and the overall error self-detection module.

Images