Unsupervised dimension-reduction online monitoring method for abnormity of secondary circuit of voltage transformer

Abstract

The invention provides an unsupervised dimensionality-reduction online monitoring method for the abnormity of a secondary circuit of a voltage transformer, and the method comprises the steps: collecting the time sequence voltage data of the voltage transformer in real time, segmenting the data according to a time window, obtaining a time-domain and frequency-domain high-dimensional data sample set in each time window, giving consideration to the time-domain characteristics of the data and the frequency-domain characteristics under an abnormal condition, and achieving the online monitoring of the abnormity of the secondary circuit of the voltage transformer. According to the method, a time domain similarity matrix and a frequency domain similarity matrix are constructed, feature vectors of the similarity matrixes are extracted and dimensionality reduction is carried out, finally, normal and abnormal moment data are distinguished by using an unsupervised Gaussian mixture model method, normal and abnormal transformer secondary circuits can be accurately distinguished by using KL divergence, and more accurate abnormal secondary circuit positioning is realized. And the normal operation of the power system is further ensured.

Description

Technical Field

[0001] The present invention relates to the field of transformer monitoring in power systems, and more specifically, to an unsupervised dimension reduction method for online monitoring of abnormalities in a secondary circuit of a voltage transformer. Background Art

[0002] The gateway electric energy metering device, which consists of high-voltage voltage transformers, secondary circuits, electric energy meters and other structures, is the core equipment for electric energy metering. Whether it can operate stably in the long term directly affects the stability and accuracy of electric energy metering data. At the same time, as the only data source for trade settlement, transactions and economic and technical indicators assessment between power generation companies, power grid companies and power users, the reliability of its measurement results also directly affects the fairness of electric energy trade. Therefore, the research on the metering performance and working status of gateway electric energy metering devices has great development potential.

[0003] During long-term use, the metering performance of the gateway electric energy metering device is susceptible to various factors and gradually declines. Since the structure of high-voltage transformers and electric energy meters is complex and the measurement accuracy is susceptible to external interference, the metering equipment may gradually become inaccurate, resulting in deviations in the calculation of electric energy, affecting the fairness of electric energy trade and even affecting the steady construction of the power market. For this reason, in the "Catalogue of Measuring Instruments Subject to Mandatory Management" issued by the State Administration for Market Regulation, gateway transformers and electric energy meters were included in the mandatory verification catalogue.

[0004] In recent years, with the continuous progress of online monitoring methods for gateway transformers, online monitoring equipment installed on the secondary side has become an important part of the gateway electric energy metering device of the substation. However, the transformer and the online monitoring device are connected through the secondary circuit, and the monitoring data contains the error information of the secondary circuit. When the secondary circuit has abnormal conditions such as loose terminals, the online monitoring device may miss or misjudge, which seriously affects the evaluation results of the operating status of the voltage transformer. Therefore, how to use the collected data to accurately determine whether the secondary circuit is abnormal has become an urgent problem to be solved. Solving this problem is of great significance to improving the reliability and accuracy of gateway electric energy metering devices. Summary of the invention

[0005] Aiming at the problem of inaccurate monitoring of abnormal voltage transformer secondary circuit in the prior art, the present invention provides an unsupervised dimension reduction abnormal online monitoring method of voltage transformer secondary circuit, comprising: Step 1: Collecting data based on the voltage transformer online monitoring device The voltage transformers are used to monitor the time-series voltage data online, wherein each voltage transformer includes three voltage transformers, and the time-series voltage data is divided into Different time windows are obtained to obtain the secondary circuit time domain high-dimensional data sample set under each time window , perform Fourier transform on the time domain high-dimensional data sample set to obtain a frequency domain high-dimensional data sample set

[0006] Step 2: respectively analyze the time domain high-dimensional data sample set and the frequency domain high-dimensional data sample set Perform similarity analysis to obtain the time domain similarity matrix under each time window and frequency domain similarity matrix

[0007] Step 3: respectively calculate the time domain similarity matrix and frequency domain similarity matrix Perform matrix decomposition to obtain the time domain feature vector and frequency domain eigenvector The time domain feature vector and frequency domain eigenvector Input into the dimensionality reduction algorithm to obtain the 1D time domain eigenvalues ​​under each time window and 1D frequency domain eigenvalues After performing maximum and minimum normalization processing on the 1D time domain eigenvalue and the 1D frequency domain eigenvalue, the two-dimensional eigenvector is obtained. is the 1-dimensional time domain eigenvalue after normalization, is the 1D frequency domain eigenvalue after normalization

[0008] Step 4: transform the two-dimensional feature vector Input into a mixed Gaussian model to identify the abnormal occurrence time window of the secondary circuit, wherein the mixed Gaussian model is used as an unsupervised classification model; Step 5: Calculate the abnormal time window and its before and after The time domain differential eigenvector of each voltage transformer under a normal time window is , g , take 2 The mean of the time domain differential feature vectors of all voltage transformers in a normal time window is obtained to obtain the benchmark differential feature mean vector , calculate the probability distribution of the time domain differential eigenvector of each voltage transformer in the abnormal time window Probability distribution of feature mean vector with baseline difference The KL divergence between them is used to identify the abnormal voltage transformer secondary circuit based on the KL divergence.

[0009] The present invention provides an unsupervised dimension reduction method for online monitoring of voltage transformer secondary circuit anomalies. Through hybrid feature analysis in the time-frequency domain, the time domain characteristics of the data and the wide-band, high-frequency characteristics under abnormal conditions can be taken into account. A similarity matrix is ​​constructed, feature vectors are extracted and dimension reduction is performed, and finally an unsupervised mixed Gaussian model method is used to distinguish between normal and abnormal time data. The KL divergence can be used to accurately distinguish between normal and abnormal transformer secondary circuits, and more accurate abnormal secondary circuit positioning can be achieved. The monitoring results obtained by the method of the present invention have the characteristics of being fast, accurate, and highly interpretable. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A flowchart of an unsupervised dimension reduction method for online monitoring of abnormalities in a secondary circuit of a voltage transformer provided by the present invention; Figure 2 An abnormal data monitoring result diagram provided by an embodiment of the present invention; Figure 3 A KL divergence calculation flow chart provided in an embodiment of the present invention; Figure 4-1 and Figure 4-2 A KL divergence analysis result diagram of an abnormal mutual inductor is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0011] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in 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 in 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. In addition, the technical features in the various embodiments or single embodiments provided by the present invention can be arbitrarily combined with each other to form a feasible technical solution. This combination is not subject to the constraints of the sequence of steps and / or the structural composition mode, but must be based on the ability of ordinary technicians in this field to achieve. When the combination of technical solutions is contradictory or cannot be achieved, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0012] At present, although the online monitoring device of the voltage transformer can evaluate the operating status of the voltage transformer in real time, it still has some shortcomings. One of the key issues is how to effectively monitor the abnormal conditions of the secondary circuit of the voltage transformer to achieve more accurate monitoring of the operating status of the voltage transformer. In order to solve this problem, the present invention proposes an innovative method, namely, an online monitoring method for abnormal data of the secondary circuit of the voltage transformer based on dimensionality reduction feature extraction and Gaussian mixture distribution. This method can accurately identify the time and location of abnormal data in the secondary circuit, thereby achieving a more comprehensive and reliable evaluation of the performance of the voltage transformer. This method can improve the practicality and reliability of the online monitoring device of the voltage transformer to a certain extent, and provide a strong guarantee for the stable operation of the power system.

[0013] The present invention provides an unsupervised dimension reduction method for online monitoring of voltage transformer secondary circuit abnormalities, the flow chart of which is as follows: Figure 1 As shown, the abnormality monitoring method includes: Step 1: Collecting data based on the voltage transformer online monitoring device The voltage transformers are used to monitor the time-series voltage data online, wherein each voltage transformer includes three voltage transformers, and the time-series voltage data is divided into Different time windows are obtained to obtain the secondary circuit time domain high-dimensional data sample set under each time window , perform Fourier transform on the time domain high-dimensional data sample set to obtain a frequency domain high-dimensional data sample set .

[0014] It is understandable that the voltage transformer online monitoring device is a multifunctional monitoring device, whose data acquisition unit is connected to the voltage transformer through a secondary circuit, and collects various key parameters of the voltage transformer in real time, including: the timestamp of data acquisition and the output voltage value of the secondary side of the voltage transformer.

[0015] Voltage transformer online monitoring device intervals the same time period Collect once The output of each group of mutual inductors (each group of mutual inductors consists of three mutual inductors, ABC) is Timing voltage data . For a fixed time period is the time window length, starting from the setting Start by extracting All data within the time period is segmented based on the time window to obtain a size of A sample set of high-dimensional data in the time domain ,in .

[0016] Perform Fourier transform on the data in the time domain high-dimensional data sample set within each time window, and the transformation formula is:

[0017] in is an imaginary unit, is the time index, is the frequency index, is the number of samples.

[0018] Then extract the frequency domain voltage amplitude data, the extraction formula is as follows:

[0019] in, is the frequency domain voltage amplitude data in the frequency domain high-dimensional data sample set, is the frequency domain voltage data in complex form obtained after Fourier transform, is the frequency index, It is frequency, express The real part of express The imaginary part of .

[0020] Step 2: respectively analyze the time domain high-dimensional data sample set and the frequency domain high-dimensional data sample set Perform similarity analysis to obtain the time domain similarity matrix under each time window and frequency domain similarity matrix .

[0021] It is understandable that according to step 1, the time domain high-dimensional data sample set under each time window is obtained and frequency domain high-dimensional data sample set , respectively, for the time domain high-dimensional data sample set and frequency domain high-dimensional data sample set Perform cosine similarity analysis. Specifically, calculate the time domain high-dimensional data sample set under each time window The cosine similarity between each two columns of data is obtained to obtain the time domain similarity matrix under each time window ; and calculating the frequency domain high-dimensional data sample set under each time window The cosine similarity between each two columns of data is obtained to obtain the frequency domain similarity matrix under each time window , where the time domain similarity matrix under each time window is and frequency domain similarity matrix The size is .

[0022] Among them, the calculation formula of cosine similarity is:

[0023] in represents the dot product of two vectors, and Represents vectors and vector The model, For vector and vector The angle between.

[0024] Step 3: respectively calculate the time domain similarity matrix and frequency domain similarity matrix Perform matrix decomposition to obtain the time domain feature vector and frequency domain eigenvector The time domain feature vector and frequency domain eigenvector Input into the dimensionality reduction algorithm to obtain the 1D time domain eigenvalues ​​under each time window and 1D frequency domain eigenvalues After performing maximum and minimum normalization processing on the 1D time domain eigenvalue and the 1D frequency domain eigenvalue, the two-dimensional eigenvector is obtained. is the 1-dimensional time domain eigenvalue after normalization, is the 1D frequency domain eigenvalue after normalization, .

[0025] It is understandable that the time domain similarity matrix under each time window is and frequency domain similarity matrix Perform matrix decomposition and get a size of The time domain feature vector and frequency domain eigenvector .

[0026] Among them, the time domain similarity matrix under each time window is and frequency domain similarity matrix Perform matrix decomposition to obtain the eigenvectors, including: For the time domain similarity matrix and frequency domain similarity matrix Perform eigenvalue decomposition respectively, and get the size of the eigenvalues The eigenvector of and , that is, solving the characteristic equation:

[0027] in is the similarity matrix, is the eigenvector composed of eigenvalues, It is the identity matrix of the same size as the similarity matrix.

[0028] Extract the time domain feature vector under each time window The principal component analysis (PCA), mean, kurtosis or median of the time window is used to obtain the 1-dimensional time domain feature value under each time window. , ; and extracting the frequency domain feature vector under each time window The principal component analysis, mean, kurtosis or median of the time window is used to obtain the 1D frequency domain eigenvalues ​​in each time window. , .

[0029] For each time window, the 1-dimensional time domain eigenvalue and 1D frequency domain eigenvalues Perform maximum and minimum normalization processing, the formula is:

[0030] in, is the original data, and are the minimum and maximum values ​​of the 1-dimensional time domain eigenvalues ​​or 1-dimensional frequency domain eigenvalues ​​under all time windows, respectively. are the 1D time domain eigenvalues ​​and 1D frequency domain eigenvalues ​​after maximum and minimum normalization.

[0031] The normalized 1D time domain eigenvalue is used as the horizontal coordinate, and the normalized 1D frequency domain eigenvalue data is used as the vertical coordinate to obtain the two-dimensional eigenvector .

[0032] Step 4: transform the two-dimensional feature vector The input is fed into the Gaussian mixture model to identify the abnormal time window of the secondary circuit. The Gaussian mixture model is used as an unsupervised classification model to classify the two-dimensional feature vector Classify and identify whether the

[0033] It is understandable that the two-dimensional feature vector Input to the mixed Gaussian model, the mixed Gaussian model can obtain the center of the Gaussian distribution and calculate the distance between the two-dimensional feature vector and the center of the Gaussian distribution .

[0034] The two-dimensional feature vector Input to the mixed Gaussian model, the mixed Gaussian model outputs a two-dimensional feature vector The covariance matrix of , standard deviation , based on 3 Criteria for taking threshold =3 .

[0035] The two-dimensional feature vector Distance from the center of the Gaussian distribution With threshold For comparison; if > , then the two-dimensional eigenvector The time window where the anomaly occurs is the time window where the anomaly occurs; otherwise, the two-dimensional feature vector The time window in which the secondary circuit is located is a normal time window. The mixed Gaussian model can be used to identify the abnormal time window of the secondary circuit from all time windows.

[0036] Among them, the mixed Gaussian model includes a probability density function composed of multiple Gaussian distributions, and the probability density function of each Gaussian distribution is expressed as:

[0037] in, is the probability density function, which represents the two-dimensional feature vector The probability distribution of is the number of Gaussian distributions, For the The weight coefficient of the Gaussian distribution, For the The density function of a Gaussian distribution is For the The mean of a Gaussian distribution, For the The covariance matrix of a Gaussian distribution.

[0038] Step 5: Calculate the abnormal time window and its before and after The time domain differential eigenvector of each voltage transformer under a normal time window is , g , take 2 The mean of the time domain differential feature vectors of all voltage transformers in a normal time window is obtained to obtain the benchmark differential feature mean vector , calculate the probability distribution of the time domain differential eigenvector of each voltage transformer in the abnormal time window Probability distribution of feature mean vector with baseline difference The KL divergence between them is used to identify the abnormal secondary circuit of the voltage transformer based on the KL divergence.

[0039] It is understandable that through step 4, the abnormal time window can be identified, and the abnormal time window and its before and after time windows can be obtained. The time domain high-dimensional sample data of each voltage transformer in a normal time window. The time domain high-dimensional sample data of each voltage transformer in a time window is differentially processed. In each time window, the time domain high-dimensional sample data of each voltage transformer is subtracted from the previous data to obtain the difference data of the two data, and the size is The time domain differential eigenvector of each transformer , The number of time series data in each time window.

[0040] Normalize the time domain differential feature vector of each voltage transformer in all time windows. Specifically, find the benchmark differential feature mean vector And the time domain differential eigenvector of each voltage transformer in the abnormal time window The maximum and minimum values ​​of all data are the mean vector of the benchmark difference feature and the time-domain differential eigenvector of each voltage transformer Normalization is performed to obtain a normalized reference differential feature mean vector and a time domain differential feature vector of each voltage transformer.

[0041] Calculate the probability distribution of the normalized reference differential feature mean vector and the time domain differential feature vector of each voltage transformer. Specifically, Divide into left open right closed interval, and and composition( +2) data intervals, with ( +2) data intervals are used to divide the normalized reference differential feature mean vector and the time domain differential feature vector of each voltage transformer.

[0042] The probability of each interval is obtained by dividing the number of data in each interval by the total number of data in all intervals, and the probability distribution of the time domain differential eigenvector of each voltage transformer in the abnormal occurrence time window is obtained. and the probability distribution of the mean vector of the benchmark difference feature .

[0043] Calculate the probability distribution of the time domain differential eigenvector of each voltage transformer in the abnormal time window Probability distribution of feature mean vector with baseline difference The KL divergence between includes:

[0044] in, represents the probability distribution of the time-domain differential eigenvector of each voltage transformer in the abnormal occurrence time window, represents the probability distribution of the benchmark differential feature mean vector, Indicates the position of the calculated element in the distribution. for and The KL divergence between .

[0045] The probability distribution of the time domain differential eigenvector of each voltage transformer in the abnormal time window is Probability distribution of feature mean vector with baseline difference The KL divergence between For comparison, when the KL divergence exceeds the threshold , the corresponding voltage transformer is the abnormal voltage transformer secondary circuit, then find the time window of the abnormality and the voltage transformer secondary circuit where the abnormality occurs, and locate the abnormal time and position of the abnormal voltage transformer.

[0046] The method provided by the present invention is described below with reference to a specific embodiment.

[0047] The voltage amplitude data of a group of three phases collected by the online monitoring device of a substation for three consecutive months can be obtained. The amplitude data matrix : ; ; ; ; Each data point is collected once every 1 minute. If 1 hour is used as the time window, there are 60 rows of data in each time window, and a time domain high-dimensional data sample set is obtained. Perform Fourier decomposition on the time domain sample data set to obtain the frequency domain high-dimensional data sample set .

[0048] For the time domain high-dimensional data sample set and frequency domain high-dimensional data sample set , respectively, and the cosine similarity analysis is performed, and the size of each time window is The time domain similarity matrix and frequency domain similarity matrix . For each time window, the time domain similarity matrix and frequency domain similarity matrix Analyze and obtain the eigenvector of the time domain similarity matrix and the eigenvector of the frequency domain similarity matrix , respectively called time domain eigenvector and frequency domain eigenvector.

[0049] The time domain feature vector and frequency domain eigenvector Input them into the PCA algorithm in the dimensionality reduction method respectively, set n_components=1, that is, get the principal component with the highest ranking, get a 1-dimensional eigenvalue corresponding to each eigenvector, and get a 1-dimensional time domain eigenvalue and 1D frequency domain eigenvalues .

[0050] The 1D time domain eigenvalue and the 1D frequency domain eigenvalue are normalized to obtain a 1D time domain eigenvalue normalized within the range of 0-1. and 1D frequency domain eigenvalues , with the normalized 1-dimensional time domain eigenvalue as Axis, the normalized 1D frequency domain eigenvalue is Axis, get the two-dimensional feature vector of the current time window .

[0051] Use the mixed Gaussian model to establish an unsupervised classification model for two-dimensional feature vectors, set the mixed Gaussian distribution number n_components=2, that is, use two Gaussian distributions to fit the quantity, and based on 3 The criterion takes an appropriate threshold to distinguish abnormal time windows from normal time windows. The two-dimensional feature vector under a certain time window If the distance d between the center of the Gaussian distribution is greater than the threshold, the time window is an abnormal time window. If the distance d is greater than the threshold, the time window is a normal time window. The recognition results are as follows: Figure 2 As shown, the abnormal occurrence time window is obtained.

[0052] Take the first abnormal time window, calculate the abnormal time window and the nearest abnormal time window before and after the abnormal time window The time domain differential eigenvector of each voltage transformer in a time window , take the mean of all transformer time domain differential eigenvectors under the normal time window, and get Benchmark difference feature mean vector , find the baseline difference feature mean vector The maximum and minimum values ​​in the time domain differential feature vector of each transformer in the abnormal time window are normalized to obtain the normalized reference differential feature mean vector and the time domain differential feature vector of each transformer in the abnormal time window.

[0053] For the normalized benchmark differential feature mean vector and the time domain differential feature vector of each transformer in the abnormal time window, , as well as The eight left-open and right-closed intervals between the two are used as the intervals where the data is located to divide the data into three intervals. The number of data in each interval is divided by the total number of data in the three intervals as the probability in this interval. The corresponding probability distribution of the benchmark differential feature mean vector and the time domain differential feature vector of each transformer under the abnormal occurrence time window is obtained. and .calculate and The KL divergence between ,calculate and The process of KL divergence between Figure 3 ,when and The KL divergence between , then the corresponding voltage transformer secondary circuit is an abnormal voltage transformer secondary circuit, otherwise, the voltage transformer secondary circuit is a normal voltage transformer secondary circuit. The final analysis and identification results are as follows Figure 4-1 and Figure 4-2 As shown, the secondary circuit of transformer 2 is an abnormal secondary circuit of transformer, and other abnormal time windows are processed in the same way.

[0054] The present invention provides an unsupervised dimension reduction online monitoring method for voltage transformer secondary circuit anomalies. The voltage transformer online monitoring device collects time series voltage data in real time, and then divides and decomposes the data in the frequency domain according to the time window to obtain a high-dimensional data sample set in the time domain and frequency domain, thereby reducing the computational cost. The cosine similarity method is used to extract features from the data sample set in the time and frequency domains, respectively, to obtain a similarity matrix. Matrix decomposition is used to obtain the eigenvectors of the time and frequency domain similarity matrices. The time and frequency domain eigenvectors are dimensionally reduced to obtain 1-dimensional time domain and frequency domain eigenvalues ​​of each time window as the 2D eigenvector of the current time window, which can more accurately extract data features and reduce the influence of redundant information. A mixed Gaussian model is used to establish an unsupervised classification model for the 2D eigenvector, and based on 3 The criterion takes an appropriate threshold to distinguish abnormal and normal data, obtains the time window of secondary circuit abnormality, and performs KL divergence analysis based on the time domain difference characteristics to obtain the abnormal transformer secondary circuit, so that the result can be more in line with the actual application scenario and can be continuously updated, thereby improving the monitoring accuracy.

[0055] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0056] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0057] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0058] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0059] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0060] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0061] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An unsupervised dimension reduction method for online monitoring of voltage transformer secondary circuit abnormality, characterized in that: include: Step 1: Collecting data based on the voltage transformer online monitoring device The voltage transformers are used to monitor the time-series voltage data online, wherein each voltage transformer includes three voltage transformers, and the time-series voltage data is divided into Different time windows are obtained to obtain the secondary circuit time domain high-dimensional data sample set under each time window , perform Fourier transform on the time domain high-dimensional data sample set to obtain a frequency domain high-dimensional data sample set Step 2: respectively analyze the time domain high-dimensional data sample set and the frequency domain high-dimensional data sample set Perform similarity analysis to obtain the time domain similarity matrix under each time window and frequency domain similarity matrix ; Step 3: respectively calculate the time domain similarity matrix and frequency domain similarity matrix Perform matrix decomposition to obtain the time domain feature vector and frequency domain eigenvector The time domain feature vector and frequency domain eigenvector Input into the dimensionality reduction algorithm to obtain the 1D time domain eigenvalues ​​under each time window and 1D frequency domain eigenvalues After performing maximum and minimum normalization processing on the 1D time domain eigenvalue and the 1D frequency domain eigenvalue, the two-dimensional eigenvector is obtained. is the 1-dimensional time domain eigenvalue after normalization, is the 1D frequency domain eigenvalue after normalization Step 4: transform the two-dimensional feature vector Input into a mixed Gaussian model to identify the abnormal occurrence time window of the secondary circuit, wherein the mixed Gaussian model is used as an unsupervised classification model; Step 5: Calculate the abnormal time window and its before and after The time domain differential eigenvector of each voltage transformer under a normal time window is , g , take 2 The mean of the time domain differential feature vectors of all voltage transformers in a normal time window is obtained to obtain the benchmark differential feature mean vector , calculate the probability distribution of the time domain differential eigenvector of each voltage transformer in the abnormal time window Probability distribution of feature mean vector with baseline difference The KL divergence between them is used to identify the abnormal voltage transformer secondary circuit based on the KL divergence.

2. The method for online monitoring of abnormalities in the secondary circuit of a voltage transformer according to claim 1, characterized in that: In step 1, the timing voltage data is divided into Different time windows are obtained to obtain the secondary circuit time domain high-dimensional data sample set under each time window ,include: At the same time interval Collect once The output of the voltage transformer is Timing voltage data , for a fixed time period is the time window length, starting from the set start time Start by extracting All the time series voltage data within the time period completes the data segmentation based on the time window and obtains the size of A sample set of high-dimensional data in the time domain ,in .

3. The method for online monitoring of abnormalities in the secondary circuit of a voltage transformer according to claim 1, characterized in that: In step 1, the time domain high-dimensional data sample set is Fourier transformed to obtain a frequency domain high-dimensional data sample set ,include: Perform Fourier transform on the time domain high-dimensional data samples in each time window, and the transformation formula is: in, is an imaginary unit, is the time index, is the frequency index, It is frequency, is the number of samples; Extract the frequency domain voltage amplitude data, the extraction formula is as follows: in, is the frequency domain voltage amplitude data in the frequency domain high-dimensional data sample set, is the frequency domain voltage data in complex form obtained after Fourier transform, express The real part of express The imaginary part of .

4. The method for online monitoring of abnormalities in the secondary circuit of a voltage transformer according to claim 1, characterized in that: In step 2, the time domain high-dimensional data sample set is respectively and the frequency domain high-dimensional data sample set Perform similarity analysis to obtain the time domain similarity matrix under each time window and frequency domain similarity matrix ,include: Calculate the time domain high-dimensional data sample set under each time window The cosine similarity between each two columns of data is obtained to obtain the time domain similarity matrix under each time window ; And calculate the frequency domain high-dimensional data sample set under each time window The cosine similarity between each two columns of data is obtained to obtain the frequency domain similarity matrix under each time window , where the time domain similarity matrix under each time window is and frequency domain similarity matrix The size is ; The calculation formula for cosine similarity is: in represents the dot product of two vectors, and Represents vectors and vector The model, For vector and vector The angle between.

5. The method for online monitoring of abnormalities in the secondary circuit of a voltage transformer according to claim 1, characterized in that: The time domain feature vector in step 3 and frequency domain eigenvector The size is , the time domain feature vector Frequency domain eigenvector Input into the dimensionality reduction algorithm to obtain the 1D time domain eigenvalues ​​under each time window and 1D frequency domain eigenvalues ,include: Extract the time domain feature vector under each time window The principal component analysis, mean, kurtosis or median of the time window is used to obtain the 1-dimensional time domain eigenvalues ​​under each time window. , ; And extract the frequency domain feature vector under each time window The principal component analysis, mean, kurtosis or median of the time window is used to obtain the 1D frequency domain eigenvalues ​​in each time window. , .

6. The method for online monitoring of abnormalities in the secondary circuit of a voltage transformer according to claim 1, characterized in that: After the 1-dimensional time domain eigenvalue and the 1-dimensional frequency domain eigenvalue are respectively subjected to maximum and minimum normalization processing, a two-dimensional eigenvector is obtained. ,include: The 1D time domain eigenvalues ​​and 1D frequency domain eigenvalues ​​under all time windows are normalized to the maximum and minimum values ​​respectively. The formula is: in, is the original data, and are the minimum and maximum values ​​of the 1-dimensional time domain eigenvalues ​​or 1-dimensional frequency domain eigenvalues ​​under all time windows, respectively. is the 1D time domain eigenvalue and 1D frequency domain eigenvalue after maximum and minimum normalization; The normalized 1D time domain eigenvalue is used as the horizontal coordinate, and the normalized 1D frequency domain eigenvalue data is used as the vertical coordinate to obtain the two-dimensional eigenvector .

7. The method for online monitoring of abnormalities in the secondary circuit of a voltage transformer according to claim 1, characterized in that: In step 4, the two-dimensional feature vector Input to the mixed Gaussian model to identify the abnormal time window of the secondary circuit, including: The two-dimensional feature vector Input to the Gaussian mixture model and output the two-dimensional feature vector The covariance matrix of Based on the covariance matrix, the feature quantity is obtained , standard deviation , threshold =3 ; The two-dimensional feature vector Distance from the center of the Gaussian distribution With threshold Make comparisons; like > , then the two-dimensional feature vector The time window where is the abnormality occurrence time window; otherwise, the two-dimensional feature vector The time window is a normal time window; The mixed Gaussian model includes a probability density function composed of multiple Gaussian distributions, and the probability density function of each Gaussian distribution is expressed as: in, is the probability density function, which represents the two-dimensional feature vector The probability distribution of is the number of Gaussian distributions, For the The weight coefficient of the Gaussian distribution, For the The density function of a Gaussian distribution is For the The mean of a Gaussian distribution, For the The covariance matrix of a Gaussian distribution.

8. The method for online monitoring of abnormalities in the secondary circuit of a voltage transformer according to claim 1, characterized in that: Step 5: Calculate the abnormal occurrence time window and its front and back The time domain differential eigenvector of each transformer under a normal time window is ,include: The time window before and after the abnormality occurs The time domain high-dimensional sample data of each voltage transformer in each time window of a normal time window is differentiated to obtain a size of The time domain differential eigenvector of each voltage transformer , The number of time series data in each time window.

9. The method for online monitoring of abnormalities in the secondary circuit of a voltage transformer according to claim 1, characterized in that: Calculate the probability distribution of the time domain differential eigenvector of each voltage transformer in the abnormal time window Probability distribution of feature mean vector with baseline difference ,include: Find the base difference feature mean vector And the time domain differential eigenvector of each voltage transformer in the abnormal time window The maximum and minimum values ​​of all data are the mean vector of the benchmark difference feature and the time-domain differential eigenvector of each voltage transformer Normalization is performed to obtain a normalized reference differential feature mean vector and a time domain differential feature vector of each voltage transformer; Will Divide into left open right closed interval, and and composition( +2) data intervals, with ( +2) data intervals are used to divide the normalized reference differential feature mean vector and the time domain differential feature vector of each voltage transformer; The probability of each interval is obtained by dividing the number of data in each interval by the total number of data in all intervals, and the probability distribution of the time domain differential eigenvector of each voltage transformer in the abnormal occurrence time window is obtained. and the probability distribution of the mean vector of the benchmark difference feature .

10. The method for online monitoring of abnormalities in the secondary circuit of a voltage transformer according to claim 1, characterized in that: The probability distribution of the time domain differential eigenvector of each voltage transformer in the abnormal occurrence time window is calculated Probability distribution of feature mean vector with baseline difference The KL divergence between includes: in, represents the probability distribution of the time-domain differential eigenvector of each voltage transformer in the abnormal occurrence time window, represents the probability distribution of the benchmark differential feature mean vector, Indicates the position of the calculated element in the distribution. for and The KL divergence between The probability distribution of the time domain differential eigenvector of each voltage transformer in the abnormal time window is Probability distribution of feature mean vector with baseline difference The KL divergence between For comparison, when the KL divergence exceeds the threshold , the corresponding voltage transformer is the abnormal voltage transformer secondary circuit.

Images