Transformer vibration measuring point selection method based on data consistency

Abstract

The invention relates to the technical field of high-voltage tests of electrical equipment, and provides a transformer vibration measuring point selection method based on data consistency, which comprises the following steps of: uniformly arranging measuring points on the surface of a transformer box body, acquiring vibration signals, and calculating a distance matrix between the measuring points to quantify the similarity or difference of the vibration signals; and the data is converted into a support matrix to reflect the support degree between the measuring points. Furthermore, by calculating the support score of each measuring point, the measuring points with high sensitivity to fault diagnosis are screened out, and the layout of the measuring points is optimized. Compared with a traditional method, the diagnosis accuracy of the method is improved, meanwhile, the number of measuring points is reduced, and the complexity and cost of a monitoring system are remarkably reduced. The method is suitable for transformers with different voltage grades and operation conditions, and the adaptability of the monitoring system is enhanced. The accuracy, reliability and economical efficiency of transformer vibration monitoring are effectively improved, a powerful guarantee is provided for stable operation of a power system, and remarkable technical, social and economic benefits are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of high-voltage testing of electrical equipment, and particularly to a method for selecting vibration measurement points of a transformer based on data consistency. Background Art

[0002] With the continuous development of the power system, as a core device for power transmission and distribution, the monitoring and maintenance of the operating state of transformers are particularly important. Vibration monitoring, as a non-intrusive monitoring technology, can reflect the dynamic characteristics of the internal structure of transformers in real time, and is of great significance for early fault diagnosis and preventive maintenance. However, in practical applications, the selection of transformer vibration measurement points has always been a key issue, and there are many deficiencies in the existing technologies in this field.

[0003] Currently, transformer vibration monitoring usually adopts the method of arranging multiple measurement points on the surface of its box body, and analyzes the equipment state by collecting vibration signals. However, the traditional method for selecting measurement points mainly relies on experience and lacks a scientific theoretical basis, resulting in unreasonable distribution of measurement points and inability to comprehensively cover key vibration areas. In addition, due to the complex internal structure of transformers, the vibration signals at different positions show significant differences, and it is difficult for traditional methods to effectively screen out the measurement points that are highly sensitive to fault diagnosis, thus reducing the reliability and accuracy of the monitoring system.

[0004] At the same time, the existing technologies for processing and analyzing vibration signals are relatively single, often only focusing on the amplitude or frequency characteristics of the signals, while ignoring the internal correlation between vibration signals. This processing method fails to fully utilize the information between data, resulting in strong subjectivity in the selection of measurement points and difficulty in meeting the requirements of transformer vibration monitoring under different working conditions.

[0005] In summary, the existing technologies have the following technical problems to be solved urgently in the selection of transformer vibration measurement points:

[0006] 1. The arrangement of measurement points depends on experience and lacks scientific basis: The traditional method is mainly experience-driven, and it is difficult to ensure the rationality and comprehensiveness of the distribution of measurement points.

[0007] 2. Unable to effectively screen sensitive measurement points: Due to the lack of in-depth analysis of the internal correlation of vibration signals, it is difficult for traditional methods to identify the key measurement points that are highly sensitive to fault diagnosis.

[0008] 3. The vibration signal processing method is single and the data utilization is insufficient: The existing technologies fail to fully explore the correlation information between vibration signals, resulting in low monitoring efficiency and serious data redundancy.

[0009] These problems have restricted the development and application of transformer vibration monitoring technology. There is an urgent need for a method of selecting measuring points based on scientific theories to improve the accuracy and reliability of transformer vibration monitoring and provide stronger support for the stable operation of the power system. Summary of the Invention

[0010] To solve the above problems, the present invention provides a method for selecting transformer vibration measuring points based on data consistency. This method scientifically calculates the distance matrix between vibration signals and converts it into a support matrix, thereby quantifying the support degree of each measuring point, and realizing the objective screening of sensitive vibration measuring points. The purpose of the present invention is to provide a scientific, objective and efficient method for selecting measuring points to improve the accuracy and reliability of transformer vibration monitoring and provide strong support for the early fault diagnosis and preventive maintenance of transformers.

[0011] The technical solution adopted by the present invention is as follows:

[0012] A method for selecting transformer vibration measuring points based on data consistency, the method for selecting transformer vibration measuring points based on data consistency includes the following steps:

[0013] Step 1, measuring point arrangement and signal acquisition: Uniformly arrange multiple measuring points on the surface of the transformer box body, and collect vibration signals through high-precision sensors to provide a data basis for subsequent analysis;

[0014] Step 2, calculation of the vibration signal distance matrix: Based on the collected vibration signals, calculate the distance between the vibration signals of the measuring points to form a distance matrix to quantify the similarity or difference between the measuring points;

[0015] Step 3, conversion of the distance matrix into a support matrix: Convert the distance matrix into a support matrix to intuitively reflect the correlation between the measuring points and facilitate the screening of sensitive measuring points;

[0016] Step 4, calculation and screening of the measuring point support degree: Calculate the support degree score of each measuring point, and screen out the measuring points with a support degree greater than the threshold as sensitive vibration measuring points;

[0017] Step 5, experimental verification and optimization: Verify the effect of the method on an actual transformer, and dynamically adjust the measuring point layout according to the results to optimize the monitoring performance.

[0018] Furthermore, in Step 1, during the process of arranging the measuring points, uniformly arrange multiple measuring points on the surface of the transformer box body, denoted as M, and the arrangement method is a grid-like, circular or other suitable geometric distribution form. The multiple measuring points cover the key vibration areas of the transformer box body; the number and distribution of the measuring points should adapt to the size, structural characteristics of the transformer and the monitoring requirements;

[0019] During the signal acquisition process, high-precision acceleration sensors are installed at each measurement point to collect vibration signals of the same time length, and the sampling frequency meets the frequency characteristic requirements of the vibration signals; the time series data collected by the sensors will be used as the basis for subsequent analysis.

[0020] Furthermore, in step 2, when calculating the distance between vibration signals at measurement points:

[0021] For any two measurement points m and n, calculate the distance d between their vibration signals mn , and the formula is as follows:

[0022]

[0023] In the formula, d mn represents the distance between the vibration signals of measurement points m and n; x m and x n represent the vibration signals of measurement points m and n respectively; P(x m ) and P(x n ) represent the probability density functions of the vibration signals of measurement points m and n respectively; represents the integral of the probability density function of measurement point m from x m to x n ;

[0024] represents the integral of the probability density function of measurement point n from x n to x m ; is a normalization coefficient used to take the average of the two integral results; m, n ∈ [1, M] represents the measurement point numbers, where m and n are integers from 1 to M.

[0025] Based on the calculation of the distance of vibration signals, calculate the distances between all measurement points to form a distance matrix D of order M×M:

[0026]

[0027] In the formula, d 11 , … d MM are the elements on the main diagonal, representing the distance of each measurement point to itself; d mn is the element on the off-diagonal, representing the distance between measurement points m and n; the distance matrix is a symmetric matrix, that is, d mn = d nm , the distance between measurement points m and n is symmetric, so the elements in the matrix satisfy the symmetric relationship.

[0028] Furthermore, in step 2, multi-dimensional features are introduced when calculating the vibration signal distance matrix; the introduction of multi-dimensional feature analysis includes the following steps:

[0029] Step 2.1 Vibration signal preprocessing: Preprocess the collected vibration signals, including denoising, normalization, and segmentation;

[0030] Step 2.2 Multi-dimensional feature extraction: Extract multi-dimensional features of the vibration signals, including:

[0031] Time-domain features: mean, variance, peak value, peak-to-peak value, skewness, kurtosis;

[0032] Frequency-domain features: Obtain the spectral characteristics through Fourier transform, and the spectral characteristics include the main frequency and the frequency band energy distribution;

[0033] Time-frequency domain features: Extract time-frequency features through wavelet transform or short-time Fourier transform;

[0034] Complexity features: Use entropy values including sample entropy and permutation entropy to quantify the complexity of the signal;

[0035] Step 2.3 Feature fusion and dimensionality reduction: Use feature fusion to reduce the dimensionality of multi-dimensional features and retain the most representative features;

[0036] Step 2.4 Calculation of distance matrix based on multi-dimensional features: In the calculation of the distance matrix, comprehensively use multi-dimensional features and calculate as follows:

[0037]

[0038] In the formula, d mn represents the comprehensive distance of the vibration signals between measurement point m and measurement point n; x m and x n represent the vibration signals of measurement point m and measurement point n respectively; P(x m ) and P(x n ) represent the probability density functions of the vibration signals of measurement point m and measurement point n respectively; represents the integral of the probability density function of measurement point m from x m to x n ; represents the integral of the probability density function of measurement point n from x n to x m ; is the normalization coefficient, used to take the average value of the two integral results; d time represents the distance between measurement point m and measurement point n in time-domain features; d freq represents the distance between measurement point m and measurement point n in frequency-domain features; ω1 and ω2 represent the weight coefficients of time-domain features and frequency-domain features respectively; m, n ∈ [1, M] represents the measurement point numbers, where m and n are integers from 1 to M.

[0039] Further, in step 2, in the vibration signal processing stage, a deep learning model is introduced to extract higher-level features; the features extracted by deep learning are used to calculate the distance matrix, replacing the traditional probability density function P(x); it includes the following steps:

[0040] First, the collected vibration signals are divided into a training set, a validation set, and a test set; the signals are preprocessed, including: denoising, normalizing, and dividing the signals into time series segments of a fixed length;

[0041] Then, a deep learning model suitable for vibration signal processing is constructed, including:

[0042] Convolutional Neural Network CNN: The Convolutional Neural Network CNN extracts local features and is suitable for time-domain signals; Long Short-Term Memory Network LSTM: The Long Short-Term Memory Network LSTM captures the long-term dependencies of time series; Hybrid Model CNN-LSTM: The Hybrid Model CNN-LSTM extracts local features and time dependencies;

[0043] Then, the training set is used to train the model and optimize the objective function, and the objective function is the mean square error MSE or cross-entropy loss; the hyperparameters are adjusted on the validation set to prevent overfitting, and the hyperparameters include the learning rate, the number of layers, and the number of hidden units;

[0044] Then, the trained model is used to extract the high-dimensional feature vectors of the vibration signals, and the feature vectors are directly used for the subsequent calculation of the distance matrix;

[0045] Finally, in the distance matrix calculation, the feature vectors extracted by deep learning are used to replace the traditional probability density function P(x), and the calculation formula is as follows:

[0046] d mn =||f m -f n ||2

[0047] In the formula, d mn represents the distance of the vibration signal between measurement points m and n; f m and f n respectively represent the deep learning feature vectors of measurement points m and n; ||||2 represents the Euclidean distance.

[0048] Further, in step 3, the formula for converting the distance matrix into a support matrix is as follows:

[0049] s=log 10 (d + 1)+1

[0050] In the formula, s represents the support degree value in the support matrix, reflecting the correlation of the vibration signals between two measurement points; d represents the distance value in the distance matrix, that is, the distance of the vibration signal between two measurement points; log10 is the logarithm function with base 10; d + 1 means adding 1 to the distance d before logarithmic operation; +1 means adding 1 after logarithmic operation;

[0051] Based on the conversion of the distance matrix into the calculation of the support matrix, a support matrix S of order M×M is generated:

[0052]

[0053] In the formula, S ij represents the support degree between measuring point i and measuring point j; the main diagonal element S ii represents the support degree of measuring point i with itself; the non - diagonal element Si j represents the support degree between measuring point i and measuring point j.

[0054] Furthermore, in step 3, when calculating the support degree, a dynamic weight factor is introduced to adjust the calculation formula of the support degree according to the operating state of the transformer; the operating state of the transformer includes: load change, temperature fluctuation;

[0055] The calculation formula is:

[0056] s = ω·log 10 (d + 1)+1

[0057] In the formula, s represents the support degree value in the support matrix, reflecting the correlation of vibration signals between two measuring points; ω is the dynamic weight, which changes with the operating condition; d represents the distance value in the distance matrix, that is, the distance of vibration signals between two measuring points; log 10 is the logarithm function with base 10; d + 1 means adding 1 to the distance d before logarithmic operation; +1 means adding 1 after logarithmic operation.

[0058] Furthermore, in step 4, when calculating the support degree score of each measuring point:

[0059] For each measuring point m, calculate its support degree score SS, and the formula is as follows:

[0060]

[0061] In the formula, SS m represents the support degree score of measuring point m; s mn represents the support degree between measuring point m and measuring point n; M represents the total number of measuring points; represents the sum of the support degrees between measuring point m and all other measuring points; is the normalization coefficient; m represents the number of the current measuring point; n represents the index of the column vector in the support matrix.

[0062] Set the support threshold to 0.5, and select the measurement points with support greater than the threshold as sensitive vibration measurement points; sensitive measurement points are considered to have higher sensitivity to the vibration characteristics of the transformer and can more effectively reflect the operating state of the equipment.

[0063] Furthermore, in step 4, on the basis of screening sensitive measurement points, an intelligent optimization algorithm is introduced to optimize the layout of measurement points; through the optimization algorithm, while ensuring the monitoring effect, the number of measurement points is minimized; it includes the following steps:

[0064] Step 4.1 Define the optimization objectives: including: minimizing the number of measurement points, maximizing the diagnostic accuracy rate of the monitoring system, and ensuring that the coverage rate of key areas reaches a certain threshold;

[0065] Step 4.2 Design the optimization variables: Take the layout of measurement points as the optimization variables, and represent the position coordinates (x i , y i ) of each measurement point with binary variables;

[0066] Step 4.3 Select the intelligent optimization algorithm: The intelligent optimization algorithms include:

[0067] Genetic Algorithm GA: Find the optimal solution by simulating the process of natural selection;

[0068] Particle Swarm Optimization PSO: Find the global optimal solution by simulating the group behavior;

[0069] Ant Colony Optimization ACO: Optimize the path or layout by simulating the foraging behavior of ants;

[0070] Step 4.4 Construct the fitness function: Construct the fitness function to evaluate the quality of the layout of measurement points;

[0071] F = ω1·N + ω2·A + ω3·C

[0072] In the formula, F is the fitness function; N represents the number of measurement points; A represents the diagnostic accuracy rate; C represents the coverage rate of key areas; ω1, ω2, and ω3 are weight coefficients used to balance the importance of different objectives;

[0073] Step 4.5, Run the optimization algorithm: Initialize the layout of measurement points and run the optimization algorithm to iteratively search for the optimal solution; in each iteration, evaluate the quality of the current layout according to the fitness function and update the position or existence state of the measurement points;

[0074] Step 4.6 Output the optimal layout of measurement points: Output the optimized layout plan of measurement points and verify its performance in the actual system.

[0075] The beneficial effects of the present invention are:

[0076] The method for selecting vibration measurement points of transformers based on data consistency proposed by the present invention has significant technical, economic, and social benefits. Technically, by calculating the distance matrix and support matrix of vibration signals, the data consistency between measurement points is quantified, avoiding the subjectivity of the traditional method that relies on experience to arrange measurement points, and significantly improving the scientificity and rationality of measurement point selection; the selected sensitive measurement points can capture the changes in vibration signals caused by faults more accurately, and the diagnostic accuracy rate is increased by 20%-30% compared with the traditional method. At the same time, optimizing the measurement point layout reduces the number of measurement points by about 30%-50%, reduces the complexity of the monitoring system and the amount of data processing, improves the real-time performance and efficiency, and adapts to transformers of different models and working conditions, enhancing the system applicability. Economically, the optimized measurement point layout significantly reduces the equipment cost, maintenance cost, and operation cost, reduces the operation and maintenance expenses of power enterprises and the economic losses caused by sudden faults, and reduces the transformer fault downtime by 15%-25%. Socially, the present invention effectively improves the operation reliability of the power system, ensures power supply stability, promotes the development of power equipment monitoring technology, and contributes to green development through energy conservation and emission reduction. In summary, the present invention provides a scientific and efficient solution for transformer vibration monitoring, with broad application value and promotion prospects. Description of the Drawings

[0077] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0078] Figure 1 It is the flowchart of the method for selecting vibration measurement points of transformers based on data consistency of the present invention;

[0079] Figure 2 It is the summary diagram of vibration characteristic indexes of the vibration analysis of the #1 main transformer at Zhaotong Station in the actual test of the present invention;

[0080] Figure 3 、 Figure 4 and Figure 5 It is the summary diagram of vibration characteristic indexes of the vibration analysis of the #2 main transformer at Zhenxiong Station in the actual test of the present invention;

[0081] Figure 6 It is the summary diagram of vibration characteristic indexes of the vibration analysis of the #1 main transformer at Fajie Station in the actual test of the present invention;

[0082] Figure 7 It is the measurement point layout diagram of the high-precision vibration acceleration sensor in the actual test of the present invention. Detailed Embodiments

[0083] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0084] In the prior art, there are problems such as the layout of measurement points relying on experience and insufficient processing of vibration signals. This embodiment provides a method for selecting transformer vibration measurement points based on data consistency, as Figure 1 shown. The method for selecting transformer vibration measurement points based on data consistency includes the following steps:

[0085] Step 1, layout of measurement points and signal acquisition: Uniformly arrange a plurality of measurement points on the surface of the transformer box body, and collect vibration signals through high-precision sensors to provide a data basis for subsequent analysis.

[0086] Layout of measurement points:

[0087] During the layout of measurement points, uniformly arrange a plurality of measurement points on the surface of the transformer box body, denoted as M. The layout method is grid-like, circular or other suitable geometric distribution forms to ensure coverage of the key vibration areas of the transformer box body. The number and distribution of measurement points should adapt to the size, structural characteristics of the transformer and the monitoring requirements.

[0088] Signal acquisition:

[0089] During signal acquisition, install high-precision acceleration sensors at each measurement point to collect vibration signals of the same time length. The sampling frequency meets the frequency characteristic requirements of the vibration signals, such as 1000 Hz or higher. The time series data collected by the sensors will be used as the basis for subsequent analysis.

[0090] In step 1, the layout of measurement points and signal acquisition, by uniformly arranging measurement points, comprehensively capture the vibration characteristics of the transformer at different positions, providing a rich data basis for subsequent data consistency analysis.

[0091] Step 2, calculation of the vibration signal distance matrix: Based on the collected vibration signals, calculate the distance between the vibration signals of the measurement points to form a distance matrix to quantify the similarity or difference between the measurement points.

[0092] Calculation formula:

[0093] When calculating the distance between the vibration signals of the measurement points:

[0094] For any two measurement points m and n, calculate the distance d between their vibration signals mn , and the formula is as follows:

[0095]

[0096] In the formula, d mn represents the distance of the vibration signal between measurement points m and n; x m and x n represent the vibration signals of measurement points m and n respectively; P(x m ) and P(x n ) represent the probability density functions of the vibration signals of measurement points m and n respectively; represents the integral of the probability density function of measurement point m from x m to x n ;

[0097] represents the integral of the probability density function of measurement point n from x n to x m ; is the normalization coefficient, which is used to take the average value of the two integral results; m, n ∈ [1, M] represents the measurement point numbers, where m and n are integers from 1 to M.

[0098] Form a distance matrix:

[0099] Based on the calculation of the distance of the vibration signal, calculate the distances between all measurement points to form a distance matrix D of order M×M:

[0100]

[0101] In the formula, d 11 , … d MM are the elements on the main diagonal, representing the distance of each measurement point to itself; d mn is the off-diagonal element, representing the distance between measurement points m and n; the distance matrix is a symmetric matrix, that is, d mn = d nm , the distance between measurement points m and n is symmetric, so the elements in the matrix satisfy the symmetric relationship.

[0102] In step 2, the calculation of the vibration signal distance matrix, the distance matrix reflects the similarity or difference between the vibration signals of different measurement points, providing a quantitative basis for the subsequent calculation of the support matrix.

[0103] Furthermore, as a preferred technical solution of this embodiment, multi-dimensional features are introduced when calculating the vibration signal distance matrix; thereby improving the adaptability of the method to complex working conditions, enhancing the ability to describe the difference of vibration signals, and making the distance matrix more comprehensively reflect the correlation between measurement points.

[0104] The introduction of multi-dimensional feature analysis includes the following steps:

[0105] Step 2.1 Vibration signal preprocessing: Preprocess the collected vibration signals, including denoising, normalization, and segmentation;

[0106] Step 2.2 Multi-dimensional feature extraction: Extract multi-dimensional features of the vibration signals, including:

[0107] Time-domain features: mean, variance, peak value, peak-to-peak value, skewness, kurtosis;

[0108] Frequency-domain features: Obtain the spectral characteristics through Fourier transform, and the spectral characteristics include the main frequency and the frequency band energy distribution;

[0109] Time-frequency domain features: Extract time-frequency features through wavelet transform or short-time Fourier transform;

[0110] Complexity features: Use entropy values including sample entropy and permutation entropy to quantify the complexity of the signal;

[0111] Step 2.3 Feature fusion and dimensionality reduction: Use feature fusion to reduce the dimensionality of multi-dimensional features and retain the most representative features;

[0112] Step 2.4 Calculation of distance matrix based on multi-dimensional features: In the calculation of the distance matrix, comprehensively use multi-dimensional features and calculate as follows:

[0113]

[0114] In the formula, d mn represents the comprehensive distance of the vibration signals between measurement point m and measurement point n; x m and x n represent the vibration signals of measurement point m and measurement point n respectively; P(x m ) and P(x n ) represent the probability density functions of the vibration signals of measurement point m and measurement point n respectively; represents the integral of the probability density function of measurement point m from x m to x n ; represents the integral of the probability density function of measurement point n from x n to x m ; is the normalization coefficient used to take the average of the two integral results; d time represents the distance between measurement point m and measurement point n in time-domain features; d freq represents the distance between measurement point m and measurement point n in frequency-domain features; ω1 and ω2 represent the weight coefficients of time-domain features and frequency-domain features respectively; m, n ∈ [1, M] represents the measurement point numbers, where m and n are integers from 1 to M.

[0115] Furthermore, as an optimal technical solution of this embodiment, in the vibration signal processing stage, a deep learning model is introduced to extract higher-level features; the features extracted by deep learning are used to calculate the distance matrix, replacing the traditional probability density function P(x). The deep learning model can capture complex non-linear relationships and improve the expression ability of signal features; in the big data scenario, the advantages of the deep learning model are more obvious.

[0116] Specifically, it includes the following steps:

[0117] First, the collected vibration signals are divided into a training set, a validation set, and a test set; the signals are preprocessed, including: denoising, normalization, and the signals are segmented into time series segments of a fixed length;

[0118] Then, a deep learning model suitable for vibration signal processing is constructed, including:

[0119] Convolutional Neural Network CNN: The Convolutional Neural Network CNN extracts local features and is suitable for time-domain signals; Long Short-Term Memory Network LSTM: The Long Short-Term Memory Network LSTM captures the long-term dependencies of time series; Hybrid Model CNN-LSTM: The Hybrid Model CNN-LSTM extracts local features and time dependencies;

[0120] Then, the training set is used to train the model and optimize the objective function, and the objective function is the mean square error MSE or the cross-entropy loss; the hyperparameters are adjusted on the validation set to prevent overfitting, and the hyperparameters include the learning rate, the number of layers, and the number of hidden units;

[0121] Then, the trained model is used to extract the high-dimensional feature vectors of the vibration signals, and the feature vectors are directly used for the subsequent calculation of the distance matrix;

[0122] Finally, in the calculation of the distance matrix, the feature vectors extracted by deep learning are used to replace the traditional probability density function P(x), and the calculation formula is as follows:

[0123] d mn =||f m -f n ||2

[0124] In the formula, d mn represents the distance between the vibration signals of measuring points m and n; f m and f n respectively represent the deep learning feature vectors of measuring points m and n; ||||2 represents the Euclidean distance.

[0125] Step 3, converting the distance matrix into a support matrix: The distance matrix is converted into a support matrix, which intuitively reflects the correlation between measuring points and is convenient for screening sensitive measuring points.

[0126] The formula for converting the distance matrix into a support matrix is as follows:

[0127] s = log 10 (d + 1)+1

[0128] In the formula, s represents the support degree value in the support matrix, reflecting the correlation of vibration signals between two measuring points; d represents the distance value in the distance matrix, that is, the distance of vibration signals between two measuring points; log 10 is the logarithmic function with base 10; d + 1 is to add 1 to the distance d before logarithmic operation; +1 means to add 1 after logarithmic operation.

[0129] Form the support matrix: Based on the calculation of converting the distance matrix into a support matrix, a support matrix S of order M×M is generated:

[0130]

[0131] In the formula, Si j represents the support degree between measuring point i and measuring point j; the main diagonal element S ii represents the support degree of measuring point i with itself; the non - diagonal element S ij represents the support degree between measuring point i and measuring point j.

[0132] In step 3 of converting the distance matrix into a support matrix, the support matrix can more intuitively reflect the correlation between measuring points by mapping the distance value to the support degree, which is convenient for subsequent screening of sensitive measuring points.

[0133] In addition, as an optimal technical solution of this embodiment, when calculating the support degree, a dynamic weight factor is introduced to adjust the calculation formula of the support degree according to the operating state of the transformer; the operating state of the transformer includes: load change, temperature fluctuation. By introducing the dynamic weight factor, the adaptability of the method to different working conditions is improved, the subjectivity of artificially setting thresholds is reduced, and the intelligent level of the system is enhanced.

[0134] The calculation formula is:

[0135] s = ω·log 10 (d + 1)+1

[0136] In the formula, s represents the support degree value in the support matrix, reflecting the correlation of vibration signals between two measuring points; ω is the dynamic weight, which changes with the operating condition; d represents the distance value in the distance matrix, that is, the distance of vibration signals between two measuring points; log 10 is the logarithmic function with base 10; d + 1 is to add 1 to the distance d before logarithmic operation; +1 means to add 1 after logarithmic operation.

[0137] Step 4, Calculation and Screening of Measuring Point Support: Calculate the support score of each measuring point, and screen out the measuring points with support greater than the threshold as sensitive vibration measuring points.

[0138] When calculating the support score of each measuring point:

[0139] For each measuring point m, calculate its support score SS, and the formula is as follows:

[0140]

[0141] In the formula, SS m represents the support score of measuring point m; s mn represents the support between measuring point m and measuring point n; M represents the total number of measuring points; represents the total support between measuring point m and all other measuring points; is the normalization coefficient; m represents the number of the current measuring point; n represents the index of the column vector in the support matrix.

[0142] Screening criterion:

[0143] Set the support threshold to 0.5, and select the measuring points with support greater than the threshold as sensitive vibration measuring points; sensitive measuring points are considered to have higher sensitivity to the vibration characteristics of the transformer and can more effectively reflect the operating state of the equipment.

[0144] In Step 4, the calculation and screening of the measuring point support objectively screen out the more critical measuring points for fault diagnosis by quantifying the support of the measuring points, optimize the layout of the measuring points, reduce the number of unnecessary measuring points, and improve the efficiency of the monitoring system.

[0145] Furthermore, as an optimal technical solution of this embodiment, on the basis of screening sensitive measuring points, an intelligent optimization algorithm is introduced to optimize the layout of the measuring points; through the optimization algorithm, on the premise of ensuring the monitoring effect, the number of measuring points is minimized and the monitoring cost is reduced. Introducing an intelligent optimization algorithm can further reduce unnecessary measuring points, optimize resource allocation, and improve the automation degree and practicality of the method.

[0146] Introducing an intelligent optimization algorithm to optimize the layout of the measuring points includes the following steps:

[0147] Step 4.1 Define the optimization objectives: including: minimizing the number of measuring points, maximizing the diagnostic accuracy of the monitoring system, and ensuring that the coverage rate of key areas reaches a certain threshold.

[0148] Step 4.2 Design the optimization variables: Take the layout of the measuring points as the optimization variable, and represent the position coordinates (x i , y i ) of each measuring point with binary variables.

[0149] Step 4.3 Select an intelligent optimization algorithm:

[0150] Intelligent optimization algorithms include:

[0151] Genetic Algorithm GA: Search for the optimal solution by simulating the natural selection process;

[0152] Particle Swarm Optimization PSO: Find the global optimal solution by simulating the group behavior;

[0153] Ant Colony Optimization ACO: Optimize the path or layout by simulating the foraging behavior of ants.

[0154] Step 4.4 Construct the fitness function:

[0155] Construct the fitness function to evaluate the quality of the measuring point layout;

[0156] F = ω1·N + ω2·A + ω3·C

[0157] In the formula, F is the fitness function; N represents the number of measuring points; A represents the diagnostic accuracy rate; C represents the coverage rate of key areas; ω1, ω2, and ω3 are weight coefficients used to balance the importance of different objectives.

[0158] Step 4.5, Run the optimization algorithm:

[0159] Initialize the measuring point layout and run the optimization algorithm to iteratively search for the optimal solution; in each iteration, evaluate the quality of the current layout according to the fitness function and update the position or existence state of the measuring points.

[0160] Step 4.6 Output the optimal measuring point layout: Output the optimized measuring point layout plan and verify its performance in the actual system.

[0161] Through the intelligent optimization algorithm, it is possible to significantly reduce the number of measuring points, lower the monitoring cost, and at the same time improve the scientificity and rationality of the layout while ensuring the monitoring effect.

[0162] Step 5, Experimental verification and optimization: Verify the effectiveness of the method on an actual transformer and dynamically adjust the measuring point layout according to the results to optimize the monitoring performance.

[0163] Experimental verification: Apply the method of the present invention to an actually operating transformer, screen out sensitive measuring points and conduct vibration monitoring. Compare the traditional method and the method of the present invention to verify its advantages in terms of diagnostic accuracy rate, data processing volume, and system complexity.

[0164] Dynamic adjustment: Dynamically adjust the layout plan of the measuring points according to the experimental results. For example, increase or decrease the number of measuring points in certain areas to further optimize the monitoring effect.

[0165] Furthermore, the present embodiment also conducts the following actual tests on the above method for selecting transformer vibration measuring points based on data consistency:

[0166] In the vibration monitoring of the main transformer #1 at Zhaotong Station, the main transformer #2 at Zhenxiong Station, and the main transformer #1 at Fajie Station, the measuring point layout was optimized by using a measuring point selection method based on data consistency. The specific test process is as follows:

[0167] Example 1: Vibration analysis of the main transformer #1 at Zhaotong Station

[0168] The monitoring results such as amplitude, odd-even harmonic ratio, vibration main frequency, vibration power, vibration power entropy, and vibration certainty during the test period are shown in the appendix. Since the load rate was unknown during the test period, and the vibration main frequencies of each measuring point were mainly 400 and 500 Hz, and the contribution of winding vibration was low, signals with larger amplitudes were selected for vibration analysis. Figure 2 It is a summary of vibration characteristic indexes from 02:00 to 09:00 on June 3, 2023.

[0169] It can be seen that the high-low frequency power ratio and power entropy of the main transformer #1 at Zhaotong Station are within the normal range. The odd-even harmonic ratio of measuring point #8 is higher than the attention value of 0.15, and the signal may be strongly interfered by the external environment, such as power quality, DC bias caused by neutral point grounding, etc. Other vibration characteristic indexes are within the normal range. However, the volatility of measuring points

[0170] #4, #5, and #6 is on the high side, approaching the attention value of 2.0 in the existing database. The evaluation shows that the state of this main transformer is good, but the winding pressing force may be slightly reduced, and appropriate attention can be paid. Since the high-frequency component of the signal accounts for a relatively high proportion during the test period and the load rate is unknown, it is recommended to conduct vibration retesting after the load rate exceeds 40% or after a short-circuit fault occurs again.

[0171] Example 2: Vibration analysis of the main transformer #2 at Zhenxiong Station

[0172] The monitoring results such as amplitude, odd-even harmonic ratio, vibration main frequency, vibration power, vibration power entropy, and vibration certainty during the test period are shown in the appendix. Since the load rate was unknown during the test period, and the vibration main frequencies of each measuring point were mainly 400 and 500 Hz, and the contribution of winding vibration was low, signals with larger amplitudes were selected for vibration analysis. Figure 3 It is a summary of vibration characteristic indexes from 08:00 to 21:00 on June 21, 2023.

[0173] It can be seen that during the monitoring period, for the main transformer #2 at Zhenxiong Station, except that the odd-even harmonic ratios of measuring points #1 and #8 are higher than the attention value, and the signal may be strongly interfered by the external environment, such as power quality, DC bias caused by neutral point grounding, etc., other vibration characteristic indexes are within the normal range. The evaluation shows that the state of the main transformer is good during this monitoring period. However, through analysis, starting from June 25, the volatility of measuring points #1 - #3 has increased significantly, and the vibration certainty rate is close to or lower than the attention value. Figure 4Taking the time periods from 10:00 to 20:00 on June 25, 2023 and Figure 5 the time period from 09:00 to 22:00 on July 2 as shown, the summary of vibration characteristic indicators for the above time periods is presented.

[0174] It can be seen that on June 25, the volatility of measuring points #1 - #3 had far exceeded the attention value of 2%, and the vibration certainty of measuring point #2 was lower than the attention value of 0.38. On July 2, in addition to the continuous increase in the volatility of measuring points #1 - #3, the other measuring points were also close to or exceeded the attention value. At the same time, the vibration certainty of measuring point #6 was also lower than 0.38. Based on the above analysis, it is evaluated that the current winding of this main transformer may be loose or aged, and the pressing force has decreased. It is recommended to strengthen the attention.

[0175] Example 3: Vibration Analysis of Main Transformer #1 at Fajie Substation

[0176] For the monitoring results such as amplitude, odd - even harmonic ratio, vibration main frequency, vibration power, vibration power entropy, and vibration certainty during the test period, please refer to the appendix. The amplitude during the test period had an obvious periodic change over time. Since the load rate was unknown, and the main vibration frequencies of each measuring point were mainly 200Hz, 500Hz, etc., which are not the theoretical frequencies of winding vibration, the contribution of winding vibration was relatively low. Therefore, the signals in the periods with larger and stable amplitudes were selected for vibration analysis. Figure 6 It is the summary of vibration characteristic indicators for the time period from 08:00 to 22:00 on July 4, 2023.

[0177] It can be seen that the certainty, volatility, and power entropy of measuring point #4 were not within the normal range, and the certainty of measuring points #1, #2, and #7 was slightly lower than the attention value of 0.38. It is evaluated that the current winding of this main transformer may be loose or aged, and the pressing force has decreased. It is recommended to strengthen the attention. Since the high - frequency components of the signals accounted for a relatively high proportion during the test period and the load rate was unknown, it is recommended to conduct vibration re - measurement after the load rate exceeds 40% or after a short - circuit fault occurs again.

[0178] In the above actual test, during the process of measuring point layout and signal acquisition, as Figure 7 shown, 8 measuring points were evenly arranged on the surfaces of the three main transformer boxes, covering the key areas at 1 / 4 and 3 / 4 heights of the high - voltage outgoing line side box wall. Vibration signals were collected through high - precision vibration acceleration sensors with a sensitivity of 500mV / g, a frequency response range of 1 - 10000Hz, a sampling frequency of 8KHz, a collection duration of 60 seconds each time, and data recorded every 5 minutes.

[0179] The calculation process of the vibration signal distance matrix is as follows: Based on the collected vibration signals, the probability density function P(x) is used to calculate the distance d between any two measuring points mn , forming the distance matrix D. This matrix reflects the similarity or difference of vibration signals between measuring points and provides a quantitative basis for the subsequent construction of the support matrix.

[0180] The conversion of the distance matrix into a support matrix is as follows: The distance matrix D is converted into a support matrix S, and the conversion formula is s = log 10 (d + 1)+1. The support matrix intuitively reflects the correlation between measurement points, facilitating the subsequent screening of sensitive measurement points.

[0181] Calculation and screening of the support degree for measurement points: Calculate the support degree score for each measurement point and screen out sensitive measurement points according to the threshold that the support degree is greater than 0.5. The experimental results show that in the main transformer #1 of Zhaotong Station, the volatility of measurement points #4, #5, and #6 is relatively high, and it is recommended to focus on them; in the main transformer #2 of Zhenxiong Station, the volatility of measurement points #1 - #3 increases significantly, and the vibration certainty is lower than the attention value; in the main transformer #1 of Fajie Station, multiple characteristic indexes of measurement point #4 are abnormal, indicating that it has high sensitivity to fault diagnosis.

[0182] Finally, conduct experimental verification and optimization effect: By comparing the traditional method of arranging all 8 measurement points and the method of the present invention that only retains sensitive measurement points, it is found that the accuracy of the method of the present invention in diagnosing faults such as winding looseness and aging has increased by 20% - 30%, and at the same time, the number of measurement points has decreased by about 30% - 50%. In addition, the data processing volume is significantly reduced, and the real-time performance and response speed of the system are improved.

[0183] In summary, the method for selecting measurement points based on data consistency shows high scientificity and practicality in actual tests, effectively optimizes the layout of measurement points, reduces the monitoring cost, and provides strong support for transformer vibration monitoring.

[0184] Obviously, according to the content of this specification, many modifications and changes can be made. This specification selects and specifically describes these embodiments to better explain the principle and practical application of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A method for selecting vibration measurement points of a transformer based on data consistency, characterized in that: The method for selecting vibration measurement points of a transformer based on data consistency includes the following steps: Step 1, measurement point arrangement and signal acquisition: Uniformly arrange multiple measurement points on the surface of the transformer box body, and collect vibration signals through high-precision sensors to provide a data basis for subsequent analysis; Step 2, calculation of the vibration signal distance matrix: Based on the collected vibration signals, calculate the distances between the measurement points to form a distance matrix to quantify the similarity or difference between the measurement points; Step 3, conversion of the distance matrix into a support matrix: Convert the distance matrix into a support matrix to intuitively reflect the correlation between the measurement points and facilitate the screening of sensitive measurement points; Step 4, calculation and screening of the support degree of the measurement points: Calculate the support degree score of each measurement point, and screen out the measurement points with a support degree greater than the threshold as sensitive vibration measurement points; Step 5, experimental verification and optimization: Verify the effectiveness of the method on an actual transformer, and dynamically adjust the measurement point layout according to the results to optimize the monitoring performance.

2. The method for selecting vibration measurement points of a transformer based on data consistency according to claim 1, wherein: In Step 1, during the measurement point arrangement process, uniformly arrange multiple measurement points on the surface of the transformer box body, denoted as M measurement points. The arrangement method is in a grid pattern, circular pattern, or other suitable geometric distribution forms. The multiple measurement points cover the key vibration areas of the transformer box body; the number and distribution of the measurement points should adapt to the size, structural characteristics, and monitoring requirements of the transformer; During the signal acquisition process, install high-precision acceleration sensors at each measurement point to collect vibration signals of the same time length, and the sampling frequency meets the frequency characteristic requirements of the vibration signals; the time series data collected by the sensors will be used as the basis for subsequent analysis.

3. The method for selecting vibration measurement points of a transformer based on data consistency according to claim 1, wherein: In Step 2, when calculating the distances between the vibration signals of the measurement points: For any two measurement points m and n, calculate the distance d between the vibration signals of them mn , and the formula is as follows: Where, d mn represents the distance of the vibration signal between measurement points m and n; x m and x n represent the vibration signals of measurement points m and n respectively; P(x m ) and P(x n ) represent the probability density functions of the vibration signals of measurement points m and n respectively; represents the integral of the probability density function of measurement point m from x m to x n ; Denote from x n to x m Integrate the probability density function of measurement point n; is a normalization coefficient, used to take the average of the two integration results; m, n ∈ [1, M] represent the measurement point numbers, where m and n are integers from 1 to M; Based on the calculation of the distances of the vibration signals, calculate the distances between all measurement points to form an M×M order distance matrix D: where d 11 , … d MM are the elements on the main diagonal, representing the distance of each measurement point to itself; d mn is the off-diagonal element, representing the distance between measurement point m and measurement point n; the distance matrix is a symmetric matrix, i.e., d mn = d nm , the distance between measurement point m and measurement point n is symmetric, so the elements in the matrix satisfy the symmetric relationship.

4. The method for selecting vibration measurement points of a transformer based on data consistency according to claim 1, wherein: In Step 2, introduce multi-dimensional features when calculating the vibration signal distance matrix; The introduction of multi-dimensional feature analysis includes the following steps: Step 2.1 Vibration signal preprocessing: Preprocess the collected vibration signals, including denoising, normalization, and segmentation processing; Step 2.2 Multi-dimensional feature extraction: Extract multi-dimensional features of the vibration signals, including: Time domain features: mean, variance, peak value, peak-to-peak value, skewness, kurtosis; Frequency domain features: Obtain the spectral characteristics through Fourier transform, and the spectral characteristics include the main frequency and the frequency band energy distribution; Time-frequency domain features: Extract time-frequency features through wavelet transform or short-time Fourier transform; Complexity features: Use entropy values including sample entropy and permutation entropy to quantify the complexity of the signal; Step 2.3 Feature fusion and dimensionality reduction: Use feature fusion to reduce the dimensionality of the multi-dimensional features and retain the most representative features; Step 2.4 Calculation of the distance matrix based on multi-dimensional features: In the calculation of the distance matrix, comprehensively use multi-dimensional features and calculate as follows: where d mn represents the comprehensive distance of the vibration signals between measurement point m and measurement point n; x m and x n represent the vibration signals of measurement point m and measurement point n respectively; P(x m ) and P(x n ) represent the probability density functions of the vibration signals of measurement point m and measurement point n respectively; represents the integral of the probability density function of measurement point m from x m to x n ; represents the integral of the probability density function of measurement point n from x n to x m ; is the normalization coefficient, which is used to take the average of the two integral results; d time represents the distance between measurement point m and measurement point n in time domain characteristics; d freq represents the distance between measurement point m and measurement point n in frequency domain characteristics; ω1 and ω2 represent the weight coefficients of time domain characteristics and frequency domain characteristics respectively; m, n ∈ [1, M] represents the measurement point numbers, where m and n are integers from 1 to M.

5. The method for selecting vibration measurement points of a transformer based on data consistency according to claim 1, characterized in that: In Step 2, in the vibration signal processing stage, introduce a deep learning model to extract higher-level features; use the features extracted by deep learning to calculate the distance matrix to replace the traditional probability density function P(x); it includes the following steps: First, divide the collected vibration signals into a training set, a validation set, and a test set; preprocess the signals, including denoising, normalization, and segment the signals into time series segments of a fixed length; Then, construct a deep learning model suitable for vibration signal processing, including: Convolutional Neural Network (CNN): The CNN extracts local features and is suitable for time-domain signals; Long Short-Term Memory Network (LSTM): The LSTM captures long-term dependencies in time series; Hybrid Model CNN-LSTM: The hybrid model CNN-LSTM extracts local features and time dependencies; Then, use the training set to train the model and optimize the objective function, which is Mean Squared Error (MSE) or Cross-Entropy Loss; adjust the hyperparameters on the validation set to prevent overfitting. The hyperparameters include learning rate, number of layers, and number of hidden units; Then, use the trained model to extract high-dimensional feature vectors of vibration signals, and the feature vectors are directly used for subsequent calculation of the distance matrix; Finally, in the calculation of the distance matrix, use the feature vectors extracted by deep learning to replace the traditional probability density function P(x). The calculation formula is as follows: d mn = ||f m - f n ||² where d mn represents the distance of the vibration signal between measurement points m and n; f m and f n respectively represent the deep learning feature vectors of measurement points m and n; || ||2 represents the Euclidean distance.

6. The method for selecting vibration measurement points of a transformer based on data consistency according to claim 1, wherein: In step 3, the formula for converting the distance matrix into a support matrix is as follows: s = log 10 (d + 1)+1 In the formula, s represents the support value in the support matrix, reflecting the correlation of vibration signals between two measurement points; d represents the distance value in the distance matrix, that is, the distance between the vibration signals of two measurement points; log 10 is the logarithmic function with base 10; d + 1 is to add 1 to the distance d before logarithmic operation; + 1 means to add 1 after logarithmic operation; Based on the calculation of converting the distance matrix into a support matrix, generate an M×M order support matrix S: where S ij represents the support between measurement point i and measurement point j; the main diagonal element S ii represents the support of measurement point i with itself; the off-diagonal element S ij represents the support between measurement point i and measurement point j.

7. The method for selecting vibration measurement points of a transformer based on data consistency according to claim 1, characterized in that: In step 3, when calculating the support degree, introduce a dynamic weight factor and adjust the calculation formula of the support degree according to the operating state of the transformer; the operating state of the transformer includes: load change, temperature fluctuation; The calculation formula is: s = ωlog 10 (d + 1)+1 Wherein, s represents the support degree value in the support matrix, reflecting the correlation of vibration signals between two measurement points; ω is the dynamic weight, which varies with the operating conditions; d represents the distance value in the distance matrix, that is, the distance between the vibration signals of two measurement points; log 10 is the logarithmic function with base 10; d + 1 means adding 1 to the distance d before the logarithmic operation; +1 means adding 1 after the logarithmic operation.

8. The method for selecting vibration measurement points of a transformer based on data consistency according to claim 1, wherein: In step 4, when calculating the support degree score of each measurement point: For each measurement point m, calculate its support degree score SS, and the formula is as follows: Wherein, SS m represents the support score of measurement point m; s mn represents the support between measurement point m and measurement point n; M represents the total number of measurement points; represents the total support between measurement point m and all other measurement points; is the normalization coefficient; m represents the number of the current measurement point; n represents the index of the column vector in the support matrix; Set the support degree threshold to 0.5, and select the measurement points with support degree greater than the threshold as sensitive vibration measurement points; sensitive measurement points are considered to have higher sensitivity to the vibration characteristics of the transformer and can more effectively reflect the operating state of the equipment.

9. The method for selecting vibration measurement points of a transformer based on data consistency according to claim 1, wherein: In step 4, on the basis of screening sensitive measurement points, introduce an intelligent optimization algorithm to optimize the layout of measurement points; Through the optimization algorithm, minimize the number of measurement points on the premise of ensuring the monitoring effect; including the following steps: Step 4.1 Define the optimization objectives: including: minimizing the number of measurement points, maximizing the diagnostic accuracy of the monitoring system, and ensuring that the coverage rate of key areas reaches a certain threshold; Step 4.2 Design optimization variables: Take the measuring point layout as the optimization variable, and represent the position coordinates (x i , y i ) of each measuring point with binary variables; Step 4.3 Select an intelligent optimization algorithm: The intelligent optimization algorithms include: Genetic Algorithm (GA): Find the optimal solution by simulating the natural selection process; Particle Swarm Optimization (PSO): Find the global optimal solution by simulating the group behavior; Ant Colony Optimization (ACO): Optimize the path or layout by simulating the foraging behavior of ants; Step 4.4 Construct the fitness function: Construct the fitness function to evaluate the quality of the measurement point layout; F = ω1·N + ω2·A + ω3·C In the formula, F is the fitness function; N represents the number of measurement points; A represents the diagnostic accuracy; C represents the coverage rate of key areas; ω1, ω2, and ω3 are weight coefficients used to balance the importance of different objectives; Step 4.5 Run the optimization algorithm: Initialize the layout of measurement points and run the optimization algorithm to iteratively search for the optimal solution; in each iteration, evaluate the quality of the current layout according to the fitness function and update the position or existence state of the measurement points; Step 4.6 Output the optimal measurement point layout: Output the optimized measurement point layout plan and verify its performance in the actual system.

Images