An adaptive identification method and system for hidden faults of a neutral line in a generator-transformer protection
By employing recursive principal component analysis and low-rank singular value decomposition techniques, an adaptive monitoring system for neutral line faults in generator-transformer units was developed. This system solved the problem of identifying irregular, weak, and latent faults such as poor neutral line contact, achieving high-sensitivity and high-accuracy fault identification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THREE GORGES JINSHAJIANG CHUANYUN HYDROPOWER DEV CO LTD
- Filing Date
- 2026-03-09
- Publication Date
- 2026-06-23
Smart Images
Figure CN122267690A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power system relay protection technology, specifically to an adaptive identification method and system for hidden faults in the neutral line of a generator-transformer unit protection system. Background Technology
[0002] In generator-transformer unit protection, abnormalities in the secondary circuit of instrument transformers can lead to malfunctions in protection systems, impacting production and grid safety. Ensuring the absence of abnormalities in the secondary circuit of instrument transformers is therefore crucial. While abnormalities in the normal phase current and phase voltage of voltage and current transformers are easily detected and identified, abnormal states of the neutral line in their secondary circuits are difficult to detect during operation and maintenance. Furthermore, abnormal states such as loose connections, open circuits, and multiple grounding points in the secondary circuit neutral line can cause malfunctions in the generator-transformer unit during external faults.
[0003] During normal operation of the generator-transformer unit, the three-phase voltage and current in the secondary circuits of the voltage and current transformers are balanced, there is no voltage difference at the neutral point, the current flowing through the neutral line is extremely weak, and the neutral line voltage is also very small, fluctuating with changes in system load. Field tests show that during normal operation, the neutral line current is below 20mA, and the voltage is in the mV range. The existing protection range and accuracy of the generator-transformer unit do not support this identification.
[0004] Besides the difficulty in sampling the neutral line, the algorithm for identifying neutral line anomalies is also a major problem. Currently, there are generally two main types of fault identification algorithms: one relies on the harmonic fault characteristics of the neutral point current and sets a harmonic amplitude threshold for judgment; the other uses neutral line fault samples to train and identify time-frequency characteristics for judgment. These two methods have achieved some results, but their application in latent faults such as poor neutral line contact is poor. Grounding and open-circuit faults caused by aging of secondary circuit lines, insulation breakdown, etc., are characterized by intermittent and irregular grounding. The harmonic amplitude and time-frequency of these latent faults have no specific characteristics and are discontinuous. Existing monitoring and judgment methods using conventional harmonic amplitude thresholds and time-frequency characteristic judgments are ineffective in identification, easily leading to omissions and false alarms.
[0005] Furthermore, the sampling data of the neutral line of the generator-transformer group will vary with the system load. Using a fixed model is obviously not widely applicable and is prone to missed and false alarms. Therefore, an adaptive model algorithm with recursive deduction is needed.
[0006] Therefore, it is necessary to design and develop an adaptive identification algorithm for irregular, weak, and hidden faults of the neutral contact type in generator-transformer groups. Summary of the Invention
[0007] To address the difficulty of detecting and identifying irregular, weak, and latent faults, particularly contact defects, in the neutral line of the secondary circuit of generator-transformer units using existing technologies, this invention employs a recursive principal component analysis (RPA) method and system tailored to the characteristics of neutral line faults in generator-transformer units. This enables adaptive monitoring of the neutral line's behavior as the system load changes. The principal component model is updated using sliding window sampling, and a low-rank singular value decomposition (LSVDe) method is used to recursively decompose the correlation matrix, enabling recursive calculation of the load matrix and eigenvalue matrix. Furthermore, exponential weighting is used to recursively update the control limits, further increasing the accuracy of abnormal state identification.
[0008] In a first aspect, the present invention provides an adaptive identification method for latent faults in the neutral line of a generator-transformer unit protection system, comprising: The neutral current and ground voltage of the voltage transformer and the neutral current, ground current and ground voltage of the current transformer of the generator-transformer group are collected by sliding window, and the data during the period of impulsive events caused by faults inside and outside the generator-transformer group are eliminated by data mutation verification to obtain a data array. The data array is standardized by zero-mean and unit-variance processing to obtain a standard dataset; Principal component analysis (PCA) algorithm is used to reduce the dimensionality of the standard dataset, extract key principal components, and construct an initial PCA model. Calculate the covariance matrix of the initial principal component analysis model, obtain the principal component direction and variance contribution, determine the number of principal components to retain, and calculate the monitoring statistical indicators and their control limits; During the system operation phase, the initial principal component analysis model is dynamically corrected by recursively updating the data of the neutral line current and ground voltage of the voltage transformer and the neutral line current, ground current and ground voltage of the current transformer in real time to obtain the recursively updated principal component analysis model. At the same time, the control limits of the statistical index are updated by recursion using exponential weighting. Based on the recursively updated principal component analysis model, the statistics of the neutral current and ground voltage of the voltage transformer and the neutral current, ground current and ground voltage of the current transformer are calculated in real time, and dynamically compared with the weighted recursively updated statistical index control limits to determine whether there is an anomaly.
[0009] In some instances, the mutation rate of the sampled data is determined by whether the mutation rate of the sampled data exceeds the mutation initiation threshold. The mutation rate of the sampled data is determined by the absolute value of the difference between the sampled value at the current moment and the sampled value of the previous week, and the absolute value of the difference between the sampled value of the previous week and the sampled value of the previous two weeks. When the mutation rate of the sampled data exceeds the mutation initiation threshold for several consecutive times, it is determined to be data from a period of shock event.
[0010] In some instances, the calculation of the covariance matrix of the initial principal component analysis model, obtaining the principal component directions and variance contributions, determining the number of principal components to retain, and calculating monitoring statistics and their control limits include: Perform principal component decomposition on the standard dataset, calculate the unbiased estimate of the covariance of the standard dataset, obtain the eigenvalues of the covariance matrix, and determine the number of principal components to retain based on the eigenvalues. Introduction Statistics and Statistics are used as monitoring statistical indicators to obtain Statistical control limits and Statistical control limits.
[0011] In some instances, the recursive update dynamically modifies the initial principal component analysis model to obtain a recursively updated principal component analysis model, while simultaneously using exponentially weighted recursive updates of statistical control limits, including: The mean, standardization matrix, and covariance matrix at time k+1 are recursively calculated using the mean, standardization matrix, and covariance matrix of the data from the first k time steps. The principal component score vector and loading vector are updated by rank-1 correction to obtain the recursively updated principal component analysis model. The control limits for statistical indicators are updated using the updated principal component analysis model.
[0012] In some instances, the real-time statistical indicators calculated by the recursively updated principal component analysis model are compared with the recursively updated control limits. If the limits are exceeded, it indicates a fault in the neutral loop; if the limits are not exceeded, it indicates that the neutral loop is operating normally, and the sampling of the recursive model and control limits continues.
[0013] Secondly, the present invention provides an adaptive identification system for latent faults in the neutral line of a generator-transformer unit protection system, comprising: The data acquisition module is used to acquire the neutral current and ground voltage of the voltage transformer and the neutral current, ground current and ground voltage of the current transformer of the generator-transformer group through sliding window acquisition, and to obtain the data array by removing the data during the period of impulsive events caused by faults inside and outside the generator-transformer group through data mutation verification. The data standardization module is used to standardize the data array by zero mean and unit variance to obtain a standard dataset. The model building module is used to perform dimensionality reduction decomposition on the standard dataset using the principal component analysis algorithm, extract key principal components, build an initial principal component analysis model, calculate the covariance matrix of the initial principal component analysis model, obtain the principal component direction and variance contribution, determine the number of principal components to be retained, and calculate the monitoring statistical indicators and their control limits. The recursive principal component analysis module is used during system operation to dynamically correct the initial principal component analysis model by recursively updating the model using real-time collected data on the neutral current and ground voltage of voltage transformers and the neutral current, ground current, and ground voltage of current transformers. At the same time, it uses exponential weighted recursive updates to control the statistical index limits. The anomaly detection module is used to calculate the statistics of the neutral current and ground voltage of the voltage transformer and the neutral current, ground current and ground voltage of the current transformer based on the recursively updated principal component analysis model, and dynamically compare them with the weighted recursively updated statistical index control limits to determine whether there is an anomaly.
[0014] In some instances, the mutation rate of the sampled data is determined by whether the mutation rate of the sampled data exceeds the mutation initiation threshold. The mutation rate of the sampled data is determined by the absolute value of the difference between the sampled value at the current moment and the sampled value of the previous week, and the absolute value of the difference between the sampled value of the previous week and the sampled value of the previous two weeks. When the mutation rate of the sampled data exceeds the mutation initiation threshold for several consecutive times, it is determined to be data from a period of shock event.
[0015] In some instances, the calculation of the covariance matrix of the initial principal component analysis model, obtaining the principal component directions and variance contributions, determining the number of principal components to retain, and calculating monitoring statistics and their control limits include: Perform principal component decomposition on the standard dataset, calculate the unbiased estimate of the covariance of the standard dataset, obtain the eigenvalues of the covariance matrix, and determine the number of principal components to retain based on the eigenvalues. Introduction Statistics and Statistics are used as monitoring statistical indicators to obtain Statistical control limits and Statistical control limits.
[0016] In some instances, the recursive update dynamically modifies the initial principal component analysis model to obtain a recursively updated principal component analysis model, while simultaneously using exponentially weighted recursive updates of statistical control limits, including: The mean, standardization matrix, and covariance matrix at time k+1 are recursively calculated using the mean, standardization matrix, and covariance matrix of the data from the first k time steps. The principal component score vector and loading vector are updated by rank-1 correction to obtain the recursively updated principal component analysis model. The control limits for statistical indicators are updated using the updated principal component analysis model.
[0017] In some instances, the real-time statistical indicators calculated by the recursively updated principal component analysis model are compared with the recursively updated control limits. If the limits are exceeded, it indicates a fault in the neutral loop; if the limits are not exceeded, it indicates that the neutral loop is operating normally, and the sampling of the recursive model and control limits continues.
[0018] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: 1. High sensitivity and strong detection capability for weak abrupt changes, improving the identification of irregular, weak, and latent faults such as poor contact. A recursive principal component analysis method is employed to address the characteristics of latent faults in the neutral line of the generator-transformer unit. The coordinates of the real-time data sampling dataset are rotated under constraints to generate new coordinates. When minor anomalies occur during the statistical process, the measurement data in the new coordinate system will show some significant deviations. By detecting the degree of this deviation, the abnormal state of the data can be assessed.
[0019] 2. It possesses adaptive capabilities. Addressing the variation in system load with the neutral line sampling data of the generator-transformer group, it utilizes low-rank singular value decomposition (LSVDe) technology to achieve online decomposition of the covariance matrix, enabling the load matrix and eigenvalue matrix to dynamically adjust over time. This effectively solves the problem of monitoring failure in time-varying processes caused by fixed models, avoids the limitations of fixed values, eliminates the need for fixed threshold configurations, and further prevents missed and false alarms of faults.
[0020] 3. It requires fewer training samples. Only a small number of initial samples are needed to build a baseline model. During subsequent runtime, sampling data is continuously collected and the covariance matrix is updated by rank-1 to achieve efficient iteration. There is no need to retrain the full amount of data, which greatly reduces the computational overhead and eliminates the dependence of conventional intelligent algorithms on a large number of training samples. Attached Figure Description
[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0022] Figure 1 This is a schematic flowchart of the adaptive identification method for hidden faults in the neutral line of the generator-transformer group protection provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of an adaptive identification system for latent faults in the neutral line of a generator-transformer unit, provided in an embodiment of the present invention. Detailed Implementation
[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and 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.
[0024] In the following description, specific embodiments of the invention will be illustrated with reference to steps and symbols performed by one or more computers, unless otherwise stated. Therefore, these steps and operations will be referred to several times as being performed by a computer, and computer execution as referred to herein includes operations by a computer processing unit representing electronic signals of data in a structured format. This operation transforms the data or maintains it at a location in the computer's memory system, which can be reconfigured or otherwise alter the operation of the computer in a manner well known to those skilled in the art. The data structure maintained by the data is the physical location of the memory, which has specific characteristics defined by the data format. However, the principles of the invention described above are not intended to be limiting, and those skilled in the art will understand that many of the following steps and operations can also be implemented in hardware.
[0025] The terms "module" or "unit" as used herein can be considered as software objects executing on the computing system. Different components, modules, engines, and services described herein can be considered as implementations on the computing system. The apparatus and methods described herein are preferably implemented in software, but can also be implemented in hardware, both of which are within the scope of this invention.
[0026] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.
[0027] In this embodiment of the invention, an adaptive identification method for latent faults in the neutral line of a generator-transformer unit is provided. This method adaptively monitors and identifies weak, irregular latent faults in the neutral line of the generator-transformer unit. Figure 1 As shown, it includes the following steps: S1. Data Acquisition: The neutral current and ground voltage of the voltage transformer and the neutral current, ground current and ground voltage of the current transformer of the generator-transformer group are acquired by sliding window acquisition, and the data during the period of impact events caused by faults inside and outside the generator-transformer group are eliminated by data mutation verification to obtain the data array. S2. Data standardization: Zero-mean and variance normalization are performed on each column of the data array to obtain a standard dataset; S3. Construct an initial principal component analysis model based on the standard dataset, calculate the covariance matrix, obtain the principal component direction and variance contribution, determine the number of principal components to be retained, and calculate the monitoring statistical indicators and their control limits. S4. Recursive principal component analysis: The principal component analysis model is recursively updated using real-time new data. S5. Recursion of statistical control limits: Update statistical indicator control limits using exponential weighting recursion. S6. Based on the recursively updated principal component analysis model, calculate the real-time data control limits, compare them with the weighted recursively updated control limits, and determine whether there is an anomaly.
[0028] Furthermore, the S1 data acquisition process is as follows: S11. Add a data acquisition unit to acquire the weak signal of the neutral line of the generator-transformer group, specifically including the neutral line current and ground voltage of the voltage transformer; the neutral line current, ground current, and ground voltage of the current transformer.
[0029] For the neutral line of the secondary circuit of the voltage transformer in the generator-transformer unit: Install a snap-on through-core current transformer (range 0-1A, accuracy 1mA) in the secondary circuit of the voltage transformer to collect the neutral line current. At the same time, the acquisition device collects the voltage between the neutral line and ground of the voltage transformer (range 0-10V, accuracy 1mV).
[0030] For the neutral line of the secondary circuit of the generator-transformer current transformer: Install snap-on through-core current transformers (range 0-1A, accuracy 1mA) in the secondary circuit and ground circuit of the current transformer to collect the neutral line current and ground current. At the same time, the acquisition device collects the neutral line-to-ground voltage of the current transformer (range 0-10V, accuracy 1mV).
[0031] S12. Sample the 5 sets of data in S11 using a sliding window at fixed time intervals, collecting m data points per window to form a data time series. A data array is formed from n consecutive time windows of data. .
[0032] S13, Mutation determination.
[0033] When a ground fault occurs inside or outside the primary system area of the generator-transformer unit, the neutral line data of the secondary circuit of the generator-transformer unit will fluctuate drastically, affecting subsequent data analysis. This data needs to be discarded. The determination method is as follows:
[0034]
[0035] Where T = cycle period, (tT) refers to the sampled value 1 week before point t, and (t-2T) refers to the sampled value 2 weeks before point t; A fixed empirical value is used as the threshold for mutation initiation. When the sampled data shows a mutation initiation exceeding the threshold four consecutive times, it is determined to be a shock event period. Data in this region is removed, and the remaining sampled data is saved to form a data array within the statistical period n. .
[0036] Furthermore, the S2 data standardization process is as follows: S21, zero-mean and variance normalization processing; A data analysis time window is formed by n consecutive time window sequences. Each data analysis interval (statistical analysis interval: each sliding window collects m data points, and n sliding windows form a statistical analysis interval) has n sets of data, and each set of data has m variables, thus forming a data array. The data array is standardized using the following formula: ,
[0037] Where E(X) is the mean of the variable and Var(X) is the variance of the variable, the standard dataset can be obtained. .
[0038] S22, score matrix and load matrix;
[0039] Standard dataset It can be decomposed into the sum of m vectors, that is:
[0040] in, Represents the score matrix; This represents the load matrix.
[0041] The vectors satisfy the following relationship:
[0042] Furthermore, S3 constructs an initial principal component analysis model, calculates the covariance matrix, obtains the principal component directions and variance contributions, determines the number of principal components to retain, calculates the monitoring statistical indicators and their control limits, and processes them as follows; S31, Principal component analysis model decomposition.
[0043] From standard dataset Multiply After performing coordinate rotation, we get:
[0044] Therefore, we can conclude that:
[0045] in,
[0046] Each component vector These are all projections of the original data, and their directions are the same as the load vector corresponding to the score vector. Reflects the standard dataset exist The projected area in the direction, and Represents standard datasets The direction of the greatest change Represents standard datasets The direction of least change.
[0047] Standard dataset after principal component decomposition It can be represented as:
[0048] in, Denotes the principal component subspace matrix; E denotes the residual subspace matrix; T denotes the principal component score matrix; P denotes the principal component loading matrix; Represents the residual score matrix; This represents the residual load matrix.
[0049] S32, Principal Component Selection.
[0050] Any two variables , correlation coefficient The unbiased estimate is expressed as:
[0051] Standard dataset The unbiased estimate of the covariance is:
[0052] according to The covariance matrix can be obtained. Eigenvalues:
[0053]
[0054] in, , Represents the eigenvalues of the covariance matrix. Represents the load vector. The larger the eigenvalue, the stronger the correlation between variables in the original data. The number of principal components to be retained is determined based on the eigenvalue. As an optional implementation method, the eigenvalue threshold is set to 85%, and those greater than this threshold are retained.
[0055] S33. Monitoring and statistical indicators and their control limits.
[0056] Introducing statistics and As a monitoring and statistical indicator.
[0057] Constructing in the principal subspace Statistics.
[0058] data The covariance matrix is:
[0059] The eigenvalue decomposition of the covariance matrix S is as follows:
[0060] in, It is a diagonal matrix. It is an orthogonal matrix and satisfies ' =I, where I is the identity matrix. Data vector projection The variance of the i-th element in the matrix is equal to the variance of the i-th element in the matrix. Let S be the i-th eigenvalue. If S is invertible, then:
[0061] The statistic is:
[0062] The statistic is the standard sum of squares of the score vector, for the sampled values of the variable at time i. ,but The statistic is:
[0063] The statistic follows an F-distribution:
[0064] in, Let F be an F-distribution with degrees of freedom k and nk.
[0065] The confidence level can be obtained as of The control limits for the statistical measure are:
[0066] when When the statistic is less than the control limit, it indicates that the sample data is normal. When the statistic exceeds the control limit, it indicates that there is an anomaly in the sample data.
[0067] The Q statistic is constructed on the residual subspace, also known as the squared prediction error (SPE).
[0068] The Q statistic can be expressed as:
[0069] in, Denotes the i-th row of the residual subspace vector E; Represents the load matrix. The residual subspace vector E is the projection of the original data X onto the residual subspace.
[0070] The control limits for SPE are:
[0071] in, , For raw data eigenvalues of the covariance matrix; The standard state distribution at the confidence level The threshold is defined as follows: k represents the number of principal components; m is the dimension of the original data.
[0072] Furthermore, the S4 recursive principal component analysis, which uses real-time acquired new data to recursively update the principal component analysis model, is as follows: As the unit operates, the neutral line data collected changes with power and load, making it a linear time-varying system. This means that the mean, variance, and covariance matrices of the neutral line collected signal are time-varying. For a time-varying linear system, using a fixed offset analysis model can lead to incorrect identification of abnormal states and false alarms. Therefore, it is necessary to collect new process data online to update the PCA analysis model. The basic principle of recursive principal component analysis is to continuously update historical data with new data.
[0073] The original data matrix is The standardized matrix is:
[0074] The mean , , It is the standard deviation of the j-th variable in the original data.
[0075] The original matrix consists of k data blocks, with the first k blocks having a length of... ,in This indicates the length of each data block collected; m represents the number of variables. The matrix is the result of standardizing k data blocks. , , Let represent the mean, standardization matrix, and covariance matrix of the first k data blocks, respectively. Using the first k time steps... , , Recursively calculate the time at time k+1 , , .
[0076] The mean of the data matrix at the first k+1 time steps can be expressed as:
[0077] Standardized as:
[0078] in,
[0079]
[0080]
[0081]
[0082]
[0083] The recursive standard deviation is:
[0084] in, This represents the i-th column of the (k+1)-th sample data. and These are the i-th elements of the corresponding vectors.
[0085] Solve for the covariance matrix using the first k+1 standardized data matrices:
[0086]
[0087] After the sample points are updated, the sample data collected in the (k+1)th time is: Then the covariance updated according to the sample points is:
[0088] The principal component score vector and loading vector are updated using the rank-1 correction method, and the covariance can be expressed as:
[0089] make , ,but:
[0090] After the first rank-1 correction, we get:
[0091]
[0092] make , .but:
[0093] After the second rank-1 correction, we get:
[0094] Therefore, the eigenvalues of the covariance matrix at this time are:
[0095] in, It is a diagonal matrix after two rank-1 corrections, and That is, the updated covariance matrix. The corresponding eigenvector.
[0096] Furthermore, the recursive update process for the S5 statistic control limits, utilizing exponential weighting, is as follows: make It is a column vector composed of the statistics of the first k-1 samples. The mean of the first k-1 statistics, It is the variance of the first k-1 statistics. When the k-th sample is detected as a normal sample, its statistic is... Join The recursive solution used for the (k+1)th control limit is as follows:
[0097]
[0098]
[0099] The weighting factor was obtained based on previous monitoring data of the neutral line in the secondary circuit of the generator-transformer unit. (less than) Introducing weighting factors have to:
[0100]
[0101] The mean and variance of the statistic obtained by exponential weighted recursion are calculated. The (k+1)th confidence level is Control limits:
[0102] in:
[0103] The final control limits are obtained by weighted fusion of the contributions of the principal component analysis model and statistics to the control limits:
[0104] in ( The fusion factor is the optimal value obtained through debugging and verification of previous data.
[0105] Therefore, it can be obtained by the control limit recursive algorithm. The (k+1)th control limit of the statistic, namely:
[0106] in, The recursive principal component analysis model determines Control limits for statistics; Each from the first k Statistical calculation.
[0107] Similarly, the (k+1)th control limit of the SPE statistic can be derived. .
[0108]
[0109] Furthermore, based on the recursively updated model, the real-time data control limits are calculated and compared with the weighted recursively updated control limits to determine whether an anomaly exists. The process is as follows: Statistic The control limits for SPE and SPE are determined by their respective sampling distributions based on a confidence level of (1-α). The control limits for the online data are calculated according to the recursively updated principal component analysis model. Statistics and The statistic is calculated in the same way as S33. Recursive calculation based on statistical control limits Limiting statistics and The limiting statistic is calculated in the same way as S5. Then, the statistical values of the online data are compared to see if they exceed the limits to monitor whether the neutral line has failed.
[0110] The calculated monitoring statistics are compared with the statistical thresholds (i.e., control limits). If the threshold is exceeded, it indicates a fault in the neutral line loop. If the threshold is not exceeded, it indicates that the neutral line loop is operating normally, and the sampling recursive model and control limits are continued.
[0111] In summary, traditional methods for identifying anomalies in the secondary circuit of generator-transformer units carry the risk of false alarms and missed alarms when dealing with irregular, weak, and latent faults such as poor contact. In contrast, the proposed method can reliably identify latent faults in the neutral line of the secondary circuit of generator-transformer units.
[0112] In another embodiment of the present invention, to facilitate better implementation of the method provided in the embodiments of the present invention, the present invention also provides a system based on the above method. The meanings of the terms are the same as in the above method, and specific implementation details can be found in the description of the method embodiments.
[0113] Please see Figure 2 , Figure 2 This is a schematic diagram of the system provided in an embodiment of the present invention. The system may include a data acquisition module 201, a data standardization processing module 202, a model building module 203, a recursive principal component analysis module 204, and an anomaly detection module 205, wherein: Data acquisition module 201 is used to acquire the neutral current and ground voltage of the voltage transformer and the neutral current, ground current and ground voltage of the current transformer of the generator-transformer group protection through a sliding window, and to obtain a data array by removing the data during the period of impulsive events caused by faults inside and outside the generator-transformer group through data mutation verification. The data standardization processing module 202 is used to perform zero-mean and unit variance standardization processing on the data array to obtain a standard dataset. The model building module 203 is used to perform dimensionality reduction decomposition on the standard dataset using the principal component analysis algorithm, extract key principal components, build an initial principal component analysis model, calculate the covariance matrix of the initial principal component analysis model, obtain the principal component direction and variance contribution, determine the number of principal components to be retained, and calculate the monitoring statistical indicators and their control limits. The recursive principal component analysis module 204 is used to dynamically correct the initial principal component analysis model during the system operation phase by using the real-time collected neutral current and ground voltage of the voltage transformer and the neutral current, ground current and ground voltage of the current transformer to obtain the recursively updated principal component analysis model. At the same time, it uses exponential weighted recursive update of statistical index control limits. The anomaly detection module 205 is used to calculate the statistics of the neutral current and ground voltage of the voltage transformer and the neutral current, ground current and ground voltage of the current transformer based on the recursively updated principal component analysis model, and dynamically compare them with the weighted recursively updated statistical index control limits to determine whether there is an anomaly.
[0114] In some instances, the mutation rate of the sampled data is determined by whether the mutation rate of the sampled data exceeds the mutation initiation threshold. The mutation rate of the sampled data is determined by the absolute value of the difference between the sampled value at the current moment and the sampled value of the previous week, and the absolute value of the difference between the sampled value of the previous week and the sampled value of the previous two weeks. When the mutation rate of the sampled data exceeds the mutation initiation threshold for several consecutive times, it is determined to be data from a period of shock event.
[0115] In some instances, the calculation of the covariance matrix of the initial principal component analysis model, obtaining the principal component directions and variance contributions, determining the number of principal components to retain, and calculating monitoring statistics and their control limits include: Perform principal component decomposition on the standard dataset, calculate the unbiased estimate of the covariance of the standard dataset, obtain the eigenvalues of the covariance matrix, and determine the number of principal components to retain based on the eigenvalues. Introduction Statistics and Statistics are used as monitoring statistical indicators to obtain Statistical control limits and Statistical control limits.
[0116] In some instances, the recursive update dynamically modifies the initial principal component analysis model to obtain a recursively updated principal component analysis model, while simultaneously using exponentially weighted recursive updates of statistical control limits, including: The mean, standardization matrix, and covariance matrix at time k+1 are recursively calculated using the mean, standardization matrix, and covariance matrix of the data from the first k time steps. The principal component score vector and loading vector are updated by rank-1 correction to obtain the recursively updated principal component analysis model. The control limits for statistical indicators are updated using the updated principal component analysis model.
[0117] In some instances, the real-time statistical indicators calculated by the recursively updated principal component analysis model are compared with the recursively updated control limits. If the limits are exceeded, it indicates a fault in the neutral loop; if the limits are not exceeded, it indicates that the neutral loop is operating normally, and the sampling of the recursive model and control limits continues.
[0118] The specific implementation methods of each module can be referred to the description of the above method embodiments, and the embodiments of the present invention will not be repeated.
[0119] The above provides a detailed description of an adaptive identification method and system for hidden faults in the neutral line of a generator-transformer unit protection system provided by the embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. An adaptive identification method for latent faults in the neutral line of a generator-transformer unit protection system, characterized in that, include: The neutral current and ground voltage of the voltage transformer and the neutral current, ground current and ground voltage of the current transformer of the generator-transformer group are collected by sliding window, and the data during the period of impulsive events caused by faults inside and outside the generator-transformer group are eliminated by data mutation verification to obtain a data array. The data array is standardized by zero-mean and unit-variance processing to obtain a standard dataset; Principal component analysis (PCA) algorithm is used to reduce the dimensionality of the standard dataset, extract key principal components, and construct an initial PCA model. Calculate the covariance matrix of the initial principal component analysis model, obtain the principal component direction and variance contribution, determine the number of principal components to retain, and calculate the monitoring statistical indicators and their control limits; During the system operation phase, the initial principal component analysis model is dynamically corrected by recursively updating the data of the neutral line current and ground voltage of the voltage transformer and the neutral line current, ground current and ground voltage of the current transformer in real time to obtain the recursively updated principal component analysis model. At the same time, the control limits of the statistical index are updated by recursion using exponential weighting. Based on the recursively updated principal component analysis model, the statistics of the neutral current and ground voltage of the voltage transformer and the neutral current, ground current and ground voltage of the current transformer are calculated in real time, and dynamically compared with the weighted recursively updated statistical index control limits to determine whether there is an anomaly.
2. The method according to claim 1, characterized in that, The data mutation amount is determined by whether the mutation amount of the sampled data is greater than the mutation initiation threshold. The mutation amount of the sampled data is determined by the absolute value of the difference between the sampled value at the current time and the sampled value of the previous week, and the absolute value of the difference between the sampled value of the previous week and the sampled value of the previous two weeks. When the mutation amount of the sampled data is greater than the mutation initiation threshold for several consecutive times, it is determined to be data from a period of shock event.
3. The method according to claim 2, characterized in that, The calculation of the covariance matrix of the initial principal component analysis model yields the principal component directions and variance contributions, determines the number of principal components to retain, and calculates monitoring statistical indicators and their control limits, including: Perform principal component decomposition on the standard dataset, calculate the unbiased estimate of the covariance of the standard dataset, obtain the eigenvalues of the covariance matrix, and determine the number of principal components to retain based on the eigenvalues. Introduction Statistics and Statistics are used as monitoring statistical indicators to obtain Statistical control limits and Statistical control limits.
4. The method according to claim 3, characterized in that, The recursive update dynamically corrects the initial principal component analysis model to obtain the recursively updated principal component analysis model, and simultaneously uses exponentially weighted recursive updates of statistical indicator control limits, including: The mean, standardization matrix, and covariance matrix at time k+1 are recursively calculated using the mean, standardization matrix, and covariance matrix of the data from the first k time steps. The principal component score vector and loading vector are updated by rank-1 correction to obtain the recursively updated principal component analysis model. The control limits for statistical indicators are updated using the updated principal component analysis model.
5. The method according to claim 1, characterized in that, The real-time statistical indicators calculated by the recursively updated principal component analysis model are compared with the recursively updated control limits. If the limits are exceeded, it indicates that there is a fault in the neutral line loop. If the limits are not exceeded, it indicates that the neutral line loop is operating normally. The sampling and recursive model and control limits are then continued.
6. An adaptive identification system for a generator-transformer unit protecting a hidden fault in the neutral line, characterized in that, include: The data acquisition module is used to acquire the neutral current and ground voltage of the voltage transformer and the neutral current, ground current and ground voltage of the current transformer of the generator-transformer group through sliding window acquisition, and to obtain the data array by removing the data during the period of impulsive events caused by faults inside and outside the generator-transformer group through data mutation verification. The data standardization module is used to standardize the data array by zero mean and unit variance to obtain a standard dataset. The model building module is used to perform dimensionality reduction decomposition on the standard dataset using the principal component analysis algorithm, extract key principal components, build an initial principal component analysis model, calculate the covariance matrix of the initial principal component analysis model, obtain the principal component direction and variance contribution, determine the number of principal components to be retained, and calculate the monitoring statistical indicators and their control limits. The recursive principal component analysis module is used during system operation to dynamically correct the initial principal component analysis model by recursively updating the model using real-time collected data on the neutral current and ground voltage of voltage transformers and the neutral current, ground current, and ground voltage of current transformers. At the same time, it uses exponential weighted recursive updates to control the statistical index limits. The anomaly detection module is used to calculate the statistics of the neutral current and ground voltage of the voltage transformer and the neutral current, ground current and ground voltage of the current transformer based on the recursively updated principal component analysis model, and dynamically compare them with the weighted recursively updated statistical index control limits to determine whether there is an anomaly.
7. The system according to claim 6, characterized in that, The data mutation amount is determined by whether the mutation amount of the sampled data is greater than the mutation initiation threshold. The mutation amount of the sampled data is determined by the absolute value of the difference between the sampled value at the current time and the sampled value of the previous week, and the absolute value of the difference between the sampled value of the previous week and the sampled value of the previous two weeks. When the mutation amount of the sampled data is greater than the mutation initiation threshold for several consecutive times, it is determined to be data from a period of shock event.
8. The system according to claim 7, characterized in that, The calculation of the covariance matrix of the initial principal component analysis model yields the principal component directions and variance contributions, determines the number of principal components to retain, and calculates monitoring statistical indicators and their control limits, including: Perform principal component decomposition on the standard dataset, calculate the unbiased estimate of the covariance of the standard dataset, obtain the eigenvalues of the covariance matrix, and determine the number of principal components to retain based on the eigenvalues. Introduction Statistics and Statistics are used as monitoring statistical indicators to obtain Statistical control limits and Statistical control limits.
9. The system according to claim 8, characterized in that, The recursive update dynamically corrects the initial principal component analysis model to obtain the recursively updated principal component analysis model, and simultaneously uses exponentially weighted recursive updates of statistical indicator control limits, including: The mean, standardization matrix, and covariance matrix at time k+1 are recursively calculated using the mean, standardization matrix, and covariance matrix of the data from the first k time steps. The principal component score vector and loading vector are updated by rank-1 correction to obtain the recursively updated principal component analysis model. The control limits for statistical indicators are updated using the updated principal component analysis model.
10. The system according to claim 6, characterized in that, The real-time statistical indicators calculated by the recursively updated principal component analysis model are compared with the recursively updated control limits. If the limits are exceeded, it indicates that there is a fault in the neutral line loop. If the limits are not exceeded, it indicates that the neutral line loop is operating normally. The sampling and recursive model and control limits are then continued.