Power transformer vibration monitoring effectiveness evaluation method and system based on multi-modal data fusion

Abstract

The application discloses a kind of based on multimodal data fusion's electric power transformer vibration monitoring efficiency evaluation method and system, belong to electric power equipment state monitoring field, this method is for the problem of insufficient quantization of existing technology to transformer box surface accessory vibration interference, the evaluation framework of multidimensional vibration evaluation index system and electric power index system is constructed in the application;Using Laplace score method eliminates redundant vibration evaluation index, and weight assignment is carried out in combination with CRITIC method;Using kernel principal component analysis (KPCA) method determines electric power characteristic index weight, finally realizes multidimensional vibration monitoring efficiency quantitative evaluation by constructing fuzzy comprehensive evaluation matrix and fuzzy synthesis operator.This method can effectively evaluate transformer vibration monitoring efficiency by the progressive analysis architecture of feature dimension reduction-weight assignment-fuzzy determination, so as to improve the reliability of transformer insulation state online monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment operation evaluation, and particularly relates to a power transformer vibration monitoring efficiency evaluation method and system based on multi-modal data fusion. BACKGROUND

[0002] The transformer vibration can directly reflect the internal mechanical state and structural characteristics of the transformer. By monitoring the vibration signal, the abnormal state in the transformer, such as winding loosening and iron core displacement, can be identified. This monitoring method helps technicians to timely discover and prevent potential faults, thereby ensuring the stable operation of the power transformer. However, the accessories installed on the surface of the box, especially the cooling devices on both sides of the box, exert a force on the transformer box when the transformer box vibrates, which causes a certain degree of influence on the transformer vibration. The vibration of the probe arranged around the accessories is different from the actual vibration of the transformer vibration source (iron core and winding coupled vibration). At the same time, the electromagnetic vibration intensity in the winding and the iron core is directly related to the load change of the transformer. When the load is high, the increase of the winding current leads to the increase of the Lorentz force, and the vibration amplitude significantly rises. In addition, the abnormal high-frequency component is introduced by the harmonic current, which interferes with the vibration spectrum characteristics. The mechanical resonance is excited by the load mutation, which affects the winding loosening or deformation. The large current of the high-voltage, medium-voltage and low-voltage side bushing easily causes the thermal expansion of the bushing conductor and the local discharge caused by the loosening of the bushing contact point, which produces the frequency multiplication component in the vibration spectrum, and causes the risk of transformer insulation deterioration.

[0003] Therefore, when carrying out the state monitoring research of the power transformer, not only the interference characteristics of the accessories (including the cooling device, the bushing current, the transformer load, etc.) on the surface of the box on the vibration signal need to be considered, but also the influence degree of these interference factors on the vibration monitoring efficiency needs to be quantitatively evaluated. By establishing the evaluation relationship model of the interference factors and the monitoring efficiency, the accuracy of the transformer state evaluation can be significantly improved.

[0004] The influence evaluation grades of the power transformer vibration monitoring efficiency include: no influence (grade I), slight influence (grade II), moderate influence (grade III) and severe influence (grade IV). The grade IV indicates that the surface accessory device of the transformer box has an influence on the state monitoring result, and the vibration source, the vibration of the accessory and the power monitoring means need to be considered when judging the insulation state of the transformer. The grade III indicates that the monitoring project can be optimized, for example, the transformer load or the iron core and the clamp current are monitored while the vibration signals of the vibration source and the accessory are monitored. The grade II indicates that the influence of the surface accessory on the insulation performance of the transformer is further reduced, and the monitoring project can be reduced, for example, the accessory power index detection is removed. The grade I indicates that only monitoring the vibration of the vibration source can effectively realize the early warning of the state of the transformer.

[0005] There is a strong correlation between the transformer insulation state and the vibration monitoring, but too many vibration evaluation indicators are not conducive to weight calculation, and also cause dimension redundancy problem, which reduces the monitoring efficiency and increases the complexity of intelligent operation and maintenance of the transformer substation, so the current evaluation method has weight distortion problem under multi-dimensional data, and the evaluation effect is not good. SUMMARY

[0006] The purpose of the present application is to provide a power transformer vibration monitoring efficiency evaluation method and system based on multi-modal data fusion, which can accurately and effectively evaluate the influence level of the transformer tank surface accessories on vibration monitoring, thereby improving the reliability of power transformer insulation state monitoring.

[0007] To achieve the above-mentioned purpose, the present application adopts a power transformer vibration monitoring efficiency evaluation method based on multi-modal data fusion, comprising:

[0008] Step S100: Collecting power transformer load data and core and clamp current data in the transformer substation to form a power evaluation index data set, collecting vibration source and accessory measurement point position coordinates and corresponding vibration data to form a multi-dimensional vibration evaluation index data set;

[0009] Step S200: Using Laplace score method to eliminate redundant vibration evaluation indicators in the multi-dimensional vibration evaluation index data set, combining CRITIC method to determine the weight of each vibration evaluation indicator, and constructing a multi-dimensional vibration evaluation index system;

[0010] Step S300: Based on the power evaluation index data set, using KPCA method to determine the weight of each power evaluation indicator, and constructing a power evaluation index system;

[0011] Step S400: Based on the multi-dimensional vibration evaluation index system and the power evaluation index system, the membership matrix of each vibration evaluation indicator and power evaluation indicator to the evaluation relationship is established in turn, and the fuzzy composition operator is used to synthesize the membership matrix and the index weight to obtain a fuzzy synthesis evaluation matrix, and the monitoring efficiency influence result is evaluated.

[0012] Further, the vibration source and accessory measurement point position coordinates in step S100 are determined by an xyz three-axis coordinate system established based on the distribution position of the power transformer accessories in the transformer substation, wherein the y-axis is along the long side direction of the transformer, the x-axis is along the short side direction of the transformer, and the z-axis is along the height direction of the transformer; the four sides of the transformer are divided into four regions A, B, C and D according to the accessory position and long and short side position, wherein the A and C regions on the long side are the surfaces where the cooling devices are located, and the B and D regions on the short side correspond to the high voltage bushing and the medium and low voltage bushing respectively; the bottom corner of the transformer A and B regions is selected as the origin O, the A and C regions are parallel to the y-z plane, and the B and D regions are parallel to the x-z plane.

[0013] Further, the accessories in step S100 include cooling devices, high-pressure sleeves, medium-pressure sleeves, and low-pressure sleeves.

[0014] Further, in step S100, the coordinates of the vibration sources and the accessory measurement points and the corresponding vibration data are collected to form a multi-dimensional vibration evaluation index data set, specifically including:

[0015] Step S101: N vibration testers are arranged on the surfaces A and C of the transformer winding and the corresponding core boxes, the vibration acceleration amplitude and phase are calculated, and the corresponding coordinates are recorded;

[0016] Step S102: The coordinates of the measurement points are combined with the vibration amplitudes, and the contour lines are divided according to the equal interval of the vibration amplitudes, a three-dimensional vibration distribution contour map is generated, and the amplitude attenuation rate between each pair of vibration measurement points is calculated , where d = 1, 2, …, N, the amplitude comparison method is used to retain M measurement points; the phase difference of the M measurement points is calculated , if and , then the coordinates of the measurement point d are the coordinates of the vibration source;

[0017] Step S103: According to the coordinates of the locations of the accessories in B and D areas and the coordinates of the vibration source, the distance between the accessories and the vibration source is calculated, and vibration tester probes are arranged on the measurement points to collect vibration data to form a multi-dimensional vibration evaluation index data set.

[0018] Further, in step S200, the Laplace score method is used to eliminate redundant vibration evaluation indexes in the multi-dimensional vibration evaluation index data set, and the CRITIC method is used to determine the weight of each vibration evaluation index, specifically including:

[0019] Step S201: The vibration evaluation index data is expanded into a two-dimensional array, which is L samples, vibration evaluation indexes, the vibration evaluation index data expanded into a two-dimensional array is normalized and formed into a matrix , where t is the total number of measurement points of the accessories and the vibration source;

[0020] Step S202: Calculate the Euclidean distance between all vibration evaluation indexes , select the appropriate k value through cross-validation method, and each vibration evaluation index selects k indexes with the closest distance to form a set ;

[0021] Step S203: The kernel norm regularization is used to calculate the low-rank adjacency matrix S, and the Laplace score matrix is obtained, and the top k vibration evaluation indexes with the highest scores are retained in descending order;

[0022] Step S204: According to the CRITIC weight formula, the standard deviation of the k vibration evaluation indexes and the correlation coefficient between the indexes are calculated to obtain the information carrying capacity of the vibration evaluation indexes , ;

[0023] Step S205: The information carrying capacity of each vibration evaluation index is normalized to obtain the weight of each vibration evaluation index ; The first to the kth weights constitute the weight set of the vibration evaluation indexes .

[0024] Further, the information carrying capacity in step S204 is:

[0025]

[0026] In the formula, is the standard deviation of the Wth vibration evaluation index, is the correlation coefficient between the Wth index and other indexes.

[0027] Further, the KPCA method is used to determine the weights of the power evaluation system indexes in step S300, specifically including:

[0028] Step S301: Standardize each power evaluation index and form a matrix , n is the number of power evaluation indexes;

[0029] Step S302: Calculate the Gaussian kernel matrix between the ith sample and the jth sample , , ;

[0030] Step S303: Center the kernel matrix K to obtain the centered kernel matrix :

[0031] Step S304: Use the eigenvalue decomposition formula to obtain the eigenvalues and eigenvectors of the covariance matrix, and calculate the variance explanation rate of as the weight value of each power evaluation index , ;

[0032] The first to the nth weight values constitute the weight set of the power evaluation indexes .

[0033] Further, the Gaussian kernel matrix is:

[0034]

[0035] wherein, is a hyperparameter of the Gaussian kernel matrix, is an exponential function;

[0036] centering the kernel matrix is:

[0037]

[0038] wherein, is a matrix with all element values , is a Gaussian kernel matrix of L samples.

[0039] Further, the step S400 specifically comprises:

[0040] sequentially calculating the inter-class distance mean of the single evaluation index of the transformer to the evaluation grade as the membership degree, and composing a membership degree matrix = performing fuzzy synthesis on the membership degree matrix and the index weight to obtain a transformer comprehensive evaluation matrix B:

[0041]

[0042] wherein, is a proportion of the vibration evaluation index system, is a fuzzy synthesis operator, is the number of evaluation grades;

[0043] the transformer comprehensive evaluation matrix B According to the maximum membership degree principle, the column where the maximum value is located is selected, that is, the transformer monitoring effectiveness evaluation grade.

[0044] A power transformer vibration monitoring effectiveness evaluation system based on multi-modal data fusion, comprising:

[0045] a data acquisition unit, which acquires power transformer load data and core and clamp current data in a substation to compose a power evaluation index data set, and acquires vibration source and accessory measurement point position coordinates and corresponding vibration data to compose a multi-dimensional vibration evaluation index data set;

[0046] a multi-dimensional vibration evaluation index system construction unit, which adopts a Laplace score method to eliminate redundant vibration evaluation indexes in the multi-dimensional vibration evaluation index data set, combines a CRITIC method to determine the weight of each vibration evaluation index, and constructs a multi-dimensional vibration evaluation index system;

[0047] The power evaluation index system construction unit constructs the power evaluation index system based on the power evaluation index data set, and determines the index weight of the power evaluation system by using the KPCA method.

[0048] The evaluation unit establishes the membership degree matrix of the evaluation relationship of each vibration evaluation index and power evaluation index in turn based on the multi-dimensional vibration evaluation index system and the power evaluation index system, and performs fuzzy synthesis on the membership degree matrix and the index weight by using a fuzzy synthesis operator to obtain a comprehensive evaluation matrix, and evaluates the monitoring effectiveness influence result.

[0049] Compared with the prior art, the power evaluation index system construction unit has the following beneficial effects:

[0050] The evaluation framework of the multi-dimensional vibration evaluation index system and the power index system is innovatively constructed, the problem of insufficient quantification of the influence of the box surface accessories on the vibration detection effectiveness is effectively solved through the collaborative analysis of the vibration signal and the electrical parameter, the weight distortion problem of the traditional method under the multi-dimensional data is solved by combining the CRITIC-KPCA combined weighting algorithm, the comparison intensity and the conflict between the indexes are objectively reflected by using the CRITIC method, the high-dimensional feature correlation is captured by using the KPCA nonlinear kernel mapping, the weight distribution is more in line with the actual mechanism, a more reliable decision basis is provided for the transformer state monitoring, the fuzzy synthesis method is used to construct the transformer comprehensive evaluation matrix, the index complexity and uncertainty can be more accurately processed, the precise quantitative evaluation of the transformer vibration monitoring effectiveness is realized, and strong technical support is provided for the intelligent operation and maintenance of the power transformer. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 It is a flowchart of the power transformer vibration monitoring effectiveness evaluation method based on multi-modal data fusion according to the present application;

[0052] Figure 2 It is a schematic diagram of the vibration measuring point position coordinate system in an embodiment of the present application;

[0053] Figure 3 It is a multi-dimensional vibration evaluation index system and power index system diagram in an embodiment of the present application. DETAILED DESCRIPTION

[0054] The present application will be further described below in combination with the drawings and embodiments.

[0055] To address the problem of insufficient quantification of vibration interference from transformer tank surface accessories in existing technologies, this invention employs the Laplace score method to effectively identify and eliminate redundant features with low contribution to condition evaluation. Combined with the CRITIC method to calculate the comparative strength and conflict between indicators, the contribution of each vibration assessment indicator to the transformer condition is objectively quantified. Electrical parameters such as transformer load and bushing currents on the high, medium, and low voltage sides mainly reflect the transmission capacity and operating status of the substation. These indicators do not have a directly observable linear relationship with the transformer vibration monitoring status. Therefore, the KPCA algorithm is used to assign weights, and a kernel function is used to map the nonlinearly correlated electrical parameters to a high-dimensional feature space, extracting key principal components and effectively quantifying electrical characteristics. This invention achieves accurate screening and weight allocation of key indicators through fuzzy synthesis of two evaluation index systems, thereby establishing a reliable transformer insulation condition vibration assessment model. Figure 1 As shown, this invention provides a method for evaluating the effectiveness of power transformer vibration monitoring based on multimodal data fusion, the steps of which include:

[0056] Step S100: Collect load data of power transformers in the substation and current data of iron core and clamps to form a power assessment index dataset; collect the location coordinates of vibration sources and accessory measuring points and the corresponding vibration data to form a multi-dimensional vibration assessment index dataset; in this embodiment, a total of 179 sample data from 9 transformers were collected.

[0057] Step S200: Use the Laplace score method to eliminate redundant vibration assessment indicators, and combine it with the CRITIC method to determine the weight of each vibration assessment indicator to construct a multidimensional vibration assessment indicator system.

[0058] Step S300: Use the KPCA method to determine the weights of each indicator in the power assessment system and construct the power assessment indicator system;

[0059] Step S400: Sequentially establish the membership degree matrix of each indicator to the evaluation fuzzy relationship. Using a fuzzy synthesis operator, fuzzily synthesize the membership degree matrix and indicator weights into a comprehensive evaluation matrix to obtain the evaluation impact level. The transformer monitoring effectiveness evaluation levels are divided into: no impact (Level I), slight impact (Level II), moderate impact (Level III), and severe impact (Level IV).

[0060] like Figure 2As shown, the acquisition of the vibration measurement point position coordinates in step S100 is based on an xyz three-axis coordinate system established according to the distribution positions of the power transformer accessories in the transformer substation, the y-axis is along the long side direction of the transformer, the x-axis is along the short side direction of the transformer, and the z-axis is along the height direction of the transformer; the four sides of the transformer are divided into four areas A, B, C and D according to the positions of the accessories and the long and short side positions, wherein the A and C areas on the long side are the surfaces where the cooling devices are located, and the B and D areas on the short side correspond to the high-voltage bushings and the medium and low-voltage bushings respectively; the bottom corners of the A and B areas of the transformer are selected as the origin O, the A and C areas are parallel to the y-z plane, and the B and D areas are parallel to the x-z plane.

[0061] Further, the specific steps of acquiring the vibration source and accessory measurement point position coordinates and the corresponding vibration data to form the multi-dimensional vibration evaluation index data set include:

[0062] Step S101: N vibration testers are arranged on the surfaces A and C of the transformer winding and the core corresponding box at the same time, the vibration acceleration amplitude , phase and corresponding coordinates are calculated, and the coordinates are recorded, wherein d=1, 2, …, N;

[0063] Step S102: the measurement point coordinates and the vibration amplitude are combined, the contour lines are divided according to the equal interval of the vibration amplitude, the three-dimensional vibration distribution contour map is generated, the amplitude attenuation rate between two vibration measurement points is calculated , and M measurement points are reserved by using the amplitude comparison method; the phase difference of the M measurement points is calculated , if and , the coordinate of the measurement point d is the coordinate of the vibration source;

[0064] Step S103: according to the coordinates of the accessory positions in the B and D areas and the coordinates of the vibration source, the vibration tester probes are arranged on the measurement points, and the vibration data is collected to form the multi-dimensional vibration evaluation index data set.

[0065] In the embodiment, the total number of the measurement points of the multi-dimensional vibration evaluation index is t=4, i.e. the positions of the vibration source, the high-voltage bushing, the low-voltage bushing and the cooling fan; the vibration parameters are the vibration acceleration effective value , the speed average value , the peak-to-peak displacement value , the main frequency , the main frequency amplitude , the odd-even harmonic ratio and the vibration entropy , =8; the power evaluation index Comprise: transformer load, core fundamental and full wave current, clamp fundamental and full wave current, n=5, as shown in Figure 3 .

[0066] The evaluation index of the excess vibration is removed by using Laplace score method, and the weight of each index is determined by combining CRITIC method, and the specific steps include:

[0067] Step S201: vibration evaluation index data is expanded into a two-dimensional array with the shape of = 179 samples index, normalized and composed into a matrix ;

[0068] Step S202: calculate the Euclidean distance between all indexes Each index is selected in turn k=10 distance nearest index set ;

[0069] Step S203: adopt the low rank adjacency matrix S of kernel norm regularization, which can be calculated by kernel norm The trace of the matrix , construct the initial adjacency matrix And perform symmetrization processing:

[0070]

[0071]

[0072] λ is the regularization parameter, after symmetrization processing, combined with the degree matrix Update the Laplace score matrix :

[0073]

[0074]

[0075] Centralized processing each vibration evaluation index adjacency matrix Get , and calculate the score , in which is a matrix with all 1s;

[0076]

[0077]

[0078] Sort in descending order to retain The top 10 indexes with the highest scores;

[0079] Step S204: According to the CRITIC weight formula, the information carrying capacity of the Wth vibration evaluation index is calculated ;

[0080]

[0081] In the formula, is the standard deviation of the Wth vibration evaluation index, is the correlation coefficient between the index W and other indexes;

[0082] Step S205: The information carrying capacity of each vibration evaluation index is normalized to obtain the weight of each vibration evaluation index ;

[0083]

[0084] Wherein, is the weight of the Wth vibration evaluation index, and the first to tenth weights constitute the weight set of the vibration evaluation index ; The 10 vibration evaluation indexes and their weight values screened in the embodiment are shown in Table 1.

[0085] Table 1: Vibration evaluation index and weight value

[0086]

[0087] Step S300, the KPCA method is used to determine the weight of each index of the power evaluation system, including steps:

[0088] Step S301: Standardize each index of the power evaluation system and form a matrix ;

[0089] Step S302: Select a Gaussian kernel, and the kernel matrix of samples and is as follows:

[0090]

[0091] Wherein, , , L is the total number of samples, is the hyperparameter of the Gaussian kernel matrix, which is set to 100, is an exponential function;

[0092] Step S303: The kernel matrix K is centered to obtain the centered kernel matrix :

[0093]

[0094] ​wherein, , is a matrix with all element values , is a Gaussian kernel matrix of L=179 samples;

[0095] Step S304: Eigen decomposition formula is used to obtain eigenvalues and eigenvectors of the covariance matrix , and the variance explanation rate of is calculated as the weight value of each power evaluation index :

[0096]

[0097] wherein, is the weight of the jth power index, and the first to fifth weights constitute a weight set of the power evaluation index ; the weight values of the five power evaluation indexes in the embodiment are shown in Table 2.

[0098] Table 2: Power evaluation index and weight value

[0099]

[0100] Step S400 of obtaining the evaluation influence grade specifically includes:

[0101] The membership degrees of the single evaluation index of the transformer to the evaluation grade are determined in sequence to to form a membership degree matrix (wherein, , =4 is the number of evaluation grades, =15), and the mean value of the distance between clusters is calculated as the membership degree; the membership degrees of the 15 indexes in the embodiment are shown in Table 3.

[0102] Table 3: Membership degree of each index

[0103]

[0104] The membership degree matrix and the index weight are subjected to fuzzy synthesis to obtain a transformer comprehensive evaluation matrix B:

[0105]

[0106] that is, wherein, is the proportion of the vibration evaluation index system, is a fuzzy synthesis operator , is the number of evaluation grades, . , ;

[0107] The evaluation influence result acquisition method is: a transformer comprehensive evaluation matrix B According to the maximum membership degree principle, the column where the maximum value is located is selected, that is, the evaluation grade.

[0108] The evaluation influence result of 34 samples of 9 transformers in the embodiment shows that 29 samples belong to no influence (level I) and 5 samples belong to moderate influence (level III); the transformer pole I low No. 1 A phase, pole I low No. 2 B phase and pole I low No. 3 B phase need to monitor the vibration of the core clamp and its accessories at the same time, and continue to observe the operation of the transformer.

[0109] The embodiment also provides a power transformer vibration monitoring effectiveness evaluation system based on multi-modal data fusion, comprising:

[0110] A data acquisition unit acquires power transformer load data and core and clamp current data in a transformer substation to form a power evaluation index data set, and acquires position coordinates of vibration sources and accessory measuring points and corresponding vibration data to form a multi-dimensional vibration evaluation index data set;

[0111] A multi-dimensional vibration evaluation index system construction unit adopts a Laplace score method to eliminate redundant vibration evaluation indexes in the multi-dimensional vibration evaluation index data set, combines a CRITIC method to determine the weight of each vibration evaluation index, and constructs a multi-dimensional vibration evaluation index system;

[0112] A power evaluation index system construction unit adopts a KPCA method to determine the weight of each index of the power evaluation system based on the power evaluation index data set, and constructs a power evaluation index system;

[0113] An evaluation unit sequentially determines the membership degree matrix of the evaluation relationship of each vibration evaluation index and power evaluation index based on the multi-dimensional vibration evaluation index system and the power evaluation index system, and performs fuzzy synthesis on the membership degree matrix and the index weight through a fuzzy synthesis operator to obtain a comprehensive evaluation matrix, and evaluates the monitoring effectiveness influence result.

[0114] It can be understood that equivalent replacement or changes of the technical solutions and inventive concepts of the present application by those skilled in the art should belong to the protection scope of the claims appended to the present application.

Claims

1. A method for evaluating the effectiveness of power transformer vibration monitoring based on multimodal data fusion, characterized in that, include: Step S100: Collect load data of power transformers in the substation and current data of iron core and clamps to form a power assessment index dataset; collect the location coordinates of vibration sources and accessory measuring points and the corresponding vibration data to form a multi-dimensional vibration assessment index dataset. Step S200: Use the Laplace score method to remove redundant vibration assessment indicators from the multidimensional vibration assessment indicator data, and combine the CRITIC method to determine the weight of each vibration assessment indicator to construct a multidimensional vibration assessment indicator system. Step S300: Based on the power assessment index dataset, use the KPCA method to determine the weight of each power assessment index and construct a power assessment index system; Step S400: Based on the multidimensional vibration assessment index system and the power assessment index system, establish the membership degree matrix of each vibration assessment index and power assessment index in turn. Through the fuzzy synthesis operator, the membership degree matrix and index weights are fuzzily synthesized into a comprehensive evaluation matrix to evaluate the impact of monitoring effectiveness.

2. The method for evaluating the effectiveness of power transformer vibration monitoring based on multimodal data fusion according to claim 1, characterized in that, In step S100, the coordinates of the vibration source and accessory measuring points are determined by a three-axis coordinate system (xyz) established based on the distribution of accessories of the power transformer within the substation. The y-axis is along the long side of the transformer, the x-axis is along the short side of the transformer, and the z-axis is along the height of the transformer. The four sides of the transformer are divided into four regions (A, B, C, and D) according to the location of the accessories and the long and short sides. Regions A and C on the long side are where the cooling devices are located, and regions B and D on the short side correspond to the high-voltage bushing and the medium- and low-voltage bushing, respectively. The bottom corners of regions A and B of the transformer are selected as the origin O. Regions A and C are parallel to the yz plane, and regions B and D are parallel to the xz plane.

3. The method for evaluating the effectiveness of power transformer vibration monitoring based on multimodal data fusion according to claim 2, characterized in that, The accessories mentioned in step S100 include a cooling device, a high-pressure bushing, a medium-pressure bushing, and a low-pressure bushing.

4. The method for evaluating the effectiveness of power transformer vibration monitoring based on multimodal data fusion according to claim 2, characterized in that, In step S100, the coordinates of the vibration source and the location of the adjacent measuring points, along with the corresponding vibration data, are collected to form a multidimensional vibration assessment index dataset, which specifically includes: Step S101: Simultaneously arrange N vibration test probes on areas A and C of the corresponding housing surface of the transformer winding and iron core, calculate the vibration acceleration amplitude and phase, and record the corresponding coordinates; Step S102: Combine the coordinates of the measuring points with the vibration amplitude, divide the vibration amplitude into contour lines at equal intervals, generate a three-dimensional vibration distribution contour map, and calculate the amplitude attenuation rate between any two vibration measuring points. Where d = 1, 2, ..., N, M measurement points are retained using the amplitude comparison method; the phase difference between the M measurement points is calculated. ,like and Then the coordinates of the measuring point d That is, the coordinates of the vibration source; Step S103: Based on the location coordinates of the attachments in areas B and D and the coordinates of the vibration source, calculate the distance between the attachments and the vibration source, and set up the vibration tester probe at the measuring point to collect vibration data and form a multidimensional vibration assessment index dataset.

5. The method for evaluating the effectiveness of power transformer vibration monitoring based on multimodal data fusion according to claim 4, characterized in that, In step S200, the Laplace score method is used to remove redundant vibration assessment indicators from the multidimensional vibration assessment indicator dataset, and the weight of each vibration assessment indicator is determined by combining the CRITIC method. Specifically, this includes: Step S201: Transfer vibration assessment index data Expanded into a two-dimensional array, containing L samples. Each vibration assessment index is normalized and formed into a matrix after being expanded into a two-dimensional array. , where t is the total number of measuring points for the attachments and vibration source; Step S202: Calculate the Euclidean distance between all vibration assessment indices. A suitable value of k is selected using cross-validation, and for each vibration assessment index, k nearest neighboring indices are selected to form a set. ; Step S203: Calculate the low-rank adjacency matrix S using nuclear norm regularization and obtain the Laplace score matrix. The k vibration assessment indicators with the highest scores are retained in descending order. Step S204: Based on the CRITIC weighting formula, calculate the standard deviation of the k vibration assessment indicators and the correlation coefficients between the indicators to obtain the information carrying capacity of the vibration assessment indicators. , ; Step S205: Normalize the information carrying capacity of each vibration assessment index to obtain the weight of each vibration assessment index. The weights from the 1st to the kth constitute the weight set of the vibration assessment index. .

6. The method for evaluating the effectiveness of power transformer vibration monitoring based on multimodal data fusion according to claim 5, characterized in that, Information carrying capacity in step S204 for: ； In the formula, Let W be the standard deviation of the Wth vibration assessment index. Let W be the correlation coefficient between indicator W and other indicators.

7. The method for evaluating the effectiveness of power transformer vibration monitoring based on multimodal data fusion according to claim 5, characterized in that, Step S300 uses the KPCA method to determine the weights of each indicator in the power assessment system, specifically including: Step S301: Standardize the various power assessment indicators and form a matrix. , where n is the number of power assessment indicators; Step S302: Calculate the i-th sample and the j-th sample Gaussian kernel matrix , , ; Step S303: Center the kernel matrix K to obtain the centered kernel matrix. : Step S304: Apply the eigenvalue decomposition formula Obtain the eigenvalues ​​of the covariance matrix and eigenvectors ,calculate The variance explained is used as the weight value for each power assessment indicator. , ; The weights from the 1st to the nth constitute the weight set of the power assessment indicators. .

8. The method for evaluating the effectiveness of power transformer vibration monitoring based on multimodal data fusion according to claim 7, characterized in that, The Gaussian kernel matrix for: ； in, Let be the hyperparameter of the Gaussian kernel matrix. It is an exponential function; Centralized kernel matrix for: ； in, All element values ​​are The matrix, Let L be the Gaussian kernel matrix for L samples.

9. The method for evaluating the effectiveness of power transformer vibration monitoring based on multimodal data fusion according to claim 7, characterized in that, Step S400 specifically includes: Calculate the mean inter-class distance of each individual evaluation index of the transformer to the evaluation level as the membership degree, and form a membership degree matrix. = The membership matrix and index weights are fuzzy synthesized to obtain the transformer comprehensive evaluation matrix B: ； in, The weighting of the vibration assessment index system, For fuzzy synthesis operators, For the evaluation level number; Transformer Comprehensive Evaluation Matrix B Based on the principle of maximum membership, the column containing the maximum value is selected as the evaluation level of transformer monitoring effectiveness.

10. A power transformer vibration monitoring performance evaluation system based on multimodal data fusion, implementing the method of any one of claims 1-9, characterized in that, include: The data acquisition unit collects load data of power transformers in the substation as well as current data of iron cores and clamps to form a power assessment index dataset. It also collects the location coordinates of vibration sources and accessory measuring points and the corresponding vibration data to form a multi-dimensional vibration assessment index dataset. The multidimensional vibration assessment index system construction unit uses the Laplace score method to remove redundant vibration assessment indexes from the multidimensional vibration assessment index data, and combines the CRITIC method to determine the weight of each vibration assessment index to construct the multidimensional vibration assessment index system. The power assessment index system construction unit uses the KPCA method to determine the weights of each index in the power assessment system based on the power assessment index dataset, and then constructs the power assessment index system. The evaluation unit, based on the multidimensional vibration assessment index system and the power assessment index system, sequentially establishes the membership degree matrix of each vibration assessment index and power assessment index to evaluate the relationship between the two. Through fuzzy synthesis operators, the membership degree matrix and index weights are fuzzily synthesized into a comprehensive evaluation matrix to evaluate the impact of monitoring effectiveness on the results.

Images