Method for detecting abnormal vibration of in-operation transformer winding

Abstract

The invention discloses an in-operation transformer winding vibration abnormity detection method, and relates to the technical field of transformer vibration state detection, and the method comprises the steps: S1, collecting a surface vibration signal during the operation of a transformer based on a vibration sensor; s2, performing winding vibration signal analysis and signal feature fusion on the surface vibration signal to obtain a winding vibration feature vector; s3, based on a time sequence prediction algorithm, taking the winding vibration feature vector as input to train a winding vibration feature prediction model; s4, predicting the winding feature vector of the to-be-detected transformer based on the trained winding vibration feature prediction model, and calculating the abnormality of the winding feature vector of the to-be-detected transformer according to a predicted value; s5, comparing the abnormality with an abnormality detection threshold value, and judging the abnormal state of the vibration of the transformer winding to be detected; the problems of poor detection effect and low efficiency in the prior art are solved, and the accuracy and the reliability of transformer winding vibration anomaly detection are remarkably improved.

Description

Technical Field

[0001] The invention relates to the technical field of transformer vibration state detection, and in particular to a method for detecting abnormal vibration of a transformer winding in operation. Background Art

[0002] Obtaining the health indicators of the transformer's internal winding operation status and performing appropriate condition maintenance or preventive maintenance on it has a significant effect. However, the traditional method of obtaining the transformer's internal winding vibration requires the construction of a finite element model and the establishment of a virtual transformer for simulation analysis, but the transformer's internal dimension information is generally missing, and due to the complexity of the transformer's internal structure and the diversity of actual working conditions, the established finite element model and simulation learning are difficult to fully and accurately reflect the real vibration behavior, thus limiting its accuracy and reliability in transformer winding status detection, resulting in poor detection effect and low detection efficiency.

[0003] Chinese patent, publication number: CN117370916A, publication date: January 9, 2024, discloses a transformer winding vibration abnormality diagnosis method, device, electronic device and medium. Obtain the current load current and the first actual vibration signal of the transformer to be tested, and calculate the first simulated vibration signal according to the current load current; perform multi-order matrix reduction projection processing on the first simulated vibration signal and the first actual vibration signal to obtain the first simulated vibration projection matrix corresponding to the first simulated vibration signal, and the first actual vibration projection matrix corresponding to the first actual vibration signal; input the first simulated vibration projection matrix into the pre-trained vibration proxy model, and output the actual vibration prediction matrix; calculate the cosine angle value between the actual vibration prediction matrix and the first actual vibration projection matrix to perform winding vibration abnormality diagnosis on the transformer to be tested; used to solve the existing problems of long calculation time, low prediction accuracy and low model generalization when performing vibration abnormality diagnosis on transformer windings. However, the invention establishes a multi-order matrix of the simulated vibration signal and the actual vibration signal of the current test transformer, establishes a finite element model to predict the projection matrix, and calculates the cosine value of the projection matrix to judge the transformer winding vibration anomaly detection method cannot fully and accurately reflect the winding vibration behavior of the transformer in actual operation. The established multi-order matrix projection may have actual deviations and cannot ensure the reliability and accuracy of the transformer winding vibration state detection. Summary of the invention

[0004] The purpose of the present invention is to address the problem that the prior art does not fully consider the complexity of the internal structure of the transformer and the diversity of actual working conditions, resulting in poor transformer winding vibration detection effect and low efficiency; a method for detecting abnormal vibration of an in-service transformer winding is proposed, by extracting winding vibration characteristics from the transformer surface vibration signal, using a time series prediction algorithm to train a transformer winding vibration characteristic prediction model with the winding vibration characteristics, predicting the transformer to be tested, calculating the abnormality degree according to the prediction error, and judging the abnormal state of the transformer winding vibration to be tested; the method overcomes the problem that the prior art does not fully consider the complexity of the internal structure of the transformer and the diversity of actual working conditions, resulting in poor transformer winding vibration detection effect and low efficiency, significantly improves the accuracy and reliability of transformer winding vibration abnormality detection, and further improves the detection efficiency.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for detecting abnormal vibration of transformer windings in operation, comprising the following steps: S1. Collecting surface vibration signals of transformer in operation based on vibration sensors; S2. Perform winding vibration signal analysis and signal feature fusion on the surface vibration signal to obtain a winding vibration feature vector; S3. Based on a time series prediction algorithm, the winding vibration feature vector is used as input to train a winding vibration feature prediction model; S4, predicting the winding characteristic vector of the transformer to be tested based on the trained winding vibration characteristic prediction model, and calculating the abnormality of the winding characteristic vector of the transformer to be tested according to the predicted value; S5. Compare the abnormality with an abnormality detection threshold to determine the abnormal state of the transformer winding vibration to be tested.

[0006] In this scheme, by analyzing and processing the original surface vibration signal of the transformer, the signal can be decomposed into inherent mode functions in different frequency bands, which effectively reduces the complexity of the original signal and has strong robustness to the unequal length, offset, amplitude change and other conditions of the time series of the original vibration signal, so as to facilitate the extraction of transformer winding vibration characteristics, which helps to comprehensively and accurately reflect the real vibration behavior; it provides strong data support for the subsequent use of time series prediction to carry out the training of the winding vibration characteristic prediction model, eliminates the influence of numerical size and dimensional difference on the prediction model training, improves the accuracy of the prediction model, and thus improves the accuracy and authenticity of the predicted value of the transformer winding vibration characteristic, ensures the accurate calculation of the abnormality, ensures the accuracy and reliability of the transformer winding vibration abnormality detection, and further improves the detection efficiency.

[0007] Preferably, S2 comprises the following sub-steps: S21, performing data cleaning and standardization processing on the surface vibration signal to obtain a surface vibration signal in a unified standard format; S22, decomposing the surface vibration signal into a plurality of signal modal components based on VMD; S23, calculating the signal modal components based on DTW to obtain a winding characteristic component data set; S24, extracting data features of the transformer historical operation data, combining the data set of the winding characteristic component with the data set to perform feature fusion, and obtaining the winding vibration feature vector.

[0008] In this scheme, the variational mode decomposition (VMD) algorithm can decompose the transformer surface vibration signal into multiple intrinsic mode functions (IMFs). These IMFs in different frequency bands respectively reflect the regularity of the sequence, effectively reducing the complexity of the sequence and improving the accuracy of subsequent model training; the dynamic time warping (DTW) algorithm can calculate the current winding correlation and voltage winding interference in the signal modal components, realize the calculation of the transformer winding vibration signal, and improve the robustness of the transformer winding vibration signal.

[0009] Preferably, the S23 includes the following sub-steps: Calculating the winding key metric of the signal modal component according to the DTW algorithm, extracting multiple data from the intersection of the winding key metric and summing them to obtain the transformer winding vibration signal; The winding characteristic component of the transformer winding vibration signal is calculated based on the transformer winding operation state, and a winding characteristic component data set of the winding characteristic component is established.

[0010] Preferably, the winding key metrics include at least current winding correlation and voltage winding interference; The signal modal component and the transformer load current are used as the first calculation sequence of the DTW algorithm, a first similarity metric between the signal modal component and the transformer load current is calculated, and the first similarity metric is used as the current winding correlation; The signal modal component and the transformer operating voltage are used as a second calculation sequence of the DTW algorithm, a second similarity metric between the signal modal component and the transformer operating voltage is calculated, and the second similarity metric is used as the voltage winding interference degree.

[0011] Preferably, the S24 includes the following sub-steps: According to the historical operation data of the transformer, the operating data characteristics of the transformer at the same time are extracted to establish a transformer operation characteristic data set; the winding characteristic data set and the transformer operation characteristic data set are merged, and the merged data set is standardized to obtain a signal feature fusion vector, and the signal feature fusion vector is used as the winding vibration feature vector.

[0012] Preferably, the winding characteristic components include at least: vibration kurtosis, average amplitude, winding maximum value, winding minimum value, winding skewness, vibration entropy, root mean square and total harmonic amplitude; The transformer operating condition data characteristics at least include: transformer operating condition factor current characteristics, transformer power characteristics and transformer oil temperature.

[0013] Preferably, S3 includes the following sub-steps: Based on the transformer winding vibration generation mechanism, the model parameters of the Informer prediction algorithm are determined through the relationship between the transformer winding vibration fundamental frequency and the load current; According to the model parameters, the winding vibration feature vector is used as the input of the Informer prediction model for model training to obtain a winding vibration feature prediction model.

[0014] Preferably, S4 includes the following sub-steps: The trained winding vibration feature prediction model is used to predict the winding vibration feature vector of the transformer to be tested, and a predicted value is obtained. The winding vibration abnormality of the transformer to be tested is calculated based on the prediction error between the predicted value and the true value.

[0015] Preferably, S5 comprises the following sub-steps: The maximum abnormality value of the vibration characteristic vector of the normal transformer winding is used as the abnormality detection threshold; Comparing the abnormality of the vibration characteristic vector of the transformer winding to be tested with the abnormality detection threshold to obtain the degree of deviation of the vibration characteristic vector of the transformer winding to be tested; The vibration state of the transformer winding to be tested is judged by the deviation degree. If the deviation degree does not exceed the standard degree threshold, the vibration state of the transformer winding to be tested is normal; if the deviation degree exceeds the standard degree threshold, the vibration state of the transformer winding to be tested is abnormal.

[0016] Preferably, the method further comprises: if the degree of deviation exceeds a standard degree threshold, classifying the abnormal state of transformer winding vibration into abnormal levels, specifically comprising: The abnormal level is divided according to the frequency and degree of exceeding the limit within a certain period of time.

[0017] Beneficial effects of the present invention: 1. By decomposing the original vibration signal and calculating the transformer winding vibration signal, the winding vibration characteristics are obtained, fully considering the complexity of the transformer internal structure and the diversity of actual working conditions, fully and accurately reflecting the real vibration situation, and ensuring the accuracy and reliability of abnormal detection of the transformer internal winding vibration state; 2. Using the time prediction algorithm to train the winding vibration feature prediction model can fully learn the winding vibration characteristics to improve the accuracy and effectiveness of the prediction model, thereby improving the accuracy and authenticity of the predicted values ​​of the transformer winding vibration characteristics, ensuring the accurate calculation of the abnormality, ensuring the accuracy and reliability of the transformer winding vibration abnormality detection, and further improving the detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Other features, objects and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings. The drawings are only for the purpose of illustrating preferred embodiments and are not to be considered as limiting the present invention. Also, the same reference symbols are used throughout the drawings to represent the same parts.

[0019] Figure 1 This is a flow chart of a method for detecting abnormal vibration of transformer windings in operation according to this embodiment.

[0020] Figure 2 This is a schematic diagram of sampling the original signal on the transformer surface in this embodiment.

[0021] Figure 3 Schematic diagram of the change of the central frequency during a VMD iteration process of this embodiment.

[0022] Figure 4 This is a schematic diagram of the result of VMD decomposition of transformer surface vibration signals in this embodiment.

[0023] Figure 5 Schematic diagram of a current winding correlation DTW calculation result of this embodiment.

[0024] Figure 6 Schematic diagram of a voltage winding interference degree DTW calculation result of this embodiment.

[0025] Figure 7 A winding vibration characteristic vector trend of this embodiment Figure 1 .

[0026] Figure 8 A winding vibration characteristic vector trend of this embodiment Figure 2 .

[0027] Fig. 9 This is a comparison diagram of winding characteristic prediction results of this embodiment.

[0028] Fig.10 Schematic diagram of the structure of a transformer winding model of this embodiment.

[0029] Fig.11 This is a schematic diagram of an Informer algorithm of this embodiment. DETAILED DESCRIPTION

[0030] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation method described herein is only an optimal embodiment of the present invention, which is only used to explain the present invention and does not limit the scope of protection of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0031] Example 1: Figure 1 As shown, a method for detecting abnormal vibration of transformer windings in operation includes steps S1-S5, wherein: S1. Collect surface vibration signals of transformer during operation based on vibration sensors.

[0032] Among them, a vibration acceleration sensor can be used to collect the original transformer surface vibration signal. For example, a single vibration acceleration sensor is selected, and the sensor technical parameter requirements are planned according to the actual signal acquisition work needs: the measuring range is vibration acceleration ±16g, the operating environment temperature is -40℃ to 85℃, the response frequency range is 5-10KHz, and the sampling resolution is 0.448mg / LSB.

[0033] As an implementation method, the vibration acceleration sensor can be installed on the low-voltage side of the transformer. The sensor measurement point is selected at the low-voltage side b phase, and the height is one-third of the ground. The vibration acceleration sensor sets its sampling frequency fs to 5KHz, the sampling time T for each sampling is 500ms, the sampling interval is 5 minutes, and continuous sampling and recording are performed. The original signal of the transformer surface is obtained as follows: Figure 2 shown.

[0034] S2. Perform winding vibration signal analysis and signal feature fusion on the surface vibration signal to obtain a winding vibration feature vector.

[0035] Specifically, S2 includes the following sub-steps: S21, performing data cleaning and standardization processing on the surface vibration signal to obtain the surface vibration signal in a unified standard format; S22, decomposing the surface vibration signal into a plurality of signal modal components based on VMD; S23, calculating the signal modal components based on DTW to obtain a winding characteristic component data set; S24. Extract data features of the transformer historical operation data, combine them with the winding characteristic component data set to perform feature fusion, and obtain the winding vibration feature vector.

[0036] Specifically, the data cleaning and standardization of the original transformer surface vibration signal include but are not limited to denoising the collected signal; wherein, the z-score (z_score) can be used for standardization. The z-score is a process in which the difference between a number and the mean is divided by the standard deviation. The method of the relative position of a certain score in the distribution can be seen, which helps to eliminate the influence of the numerical size and dimensional difference of the surface vibration signal on the model training.

[0037] Specifically, S23 includes the following sub-steps: The winding key metric of the signal modal component is calculated according to the DTW algorithm, and multiple data are extracted from the intersection of the winding key metric and summed to obtain the transformer winding vibration signal; A winding characteristic component of a transformer winding vibration signal is calculated based on the transformer winding operation state, and a winding characteristic component data set of the winding characteristic component is established.

[0038] Specifically, the winding key metrics include at least current winding correlation and voltage winding interference; Among them, the signal modal component and the transformer load current are used as the first calculation sequence of the DTW algorithm, the first similarity measure of the signal modal component and the transformer load current is calculated, and the first similarity measure is used as the current winding correlation; the signal modal component and the transformer operating voltage are used as the second calculation sequence of the DTW algorithm, the second similarity measure of the signal modal component and the transformer operating voltage is calculated, and the second similarity measure is used as the voltage winding interference.

[0039] Specifically, S24 includes the following sub-steps: According to the historical operation data of the transformer, the operating data characteristics of the transformer at the same time are extracted to establish a transformer operation feature data set; the winding feature data set and the transformer operation feature data set are merged, and the merged data set is standardized to obtain a signal feature fusion vector, and the signal feature fusion vector is used as the winding vibration feature vector.

[0040] Specifically, the winding characteristic components include at least: vibration kurtosis, average amplitude, winding maximum value, winding minimum value, winding skewness, vibration entropy, root mean square and total harmonic amplitude; The transformer operating condition data characteristics include at least: transformer operating condition factor current characteristics, transformer power characteristics and transformer oil temperature.

[0041] S3. Based on the time series prediction algorithm, the winding vibration feature vector is used as input to train the winding vibration feature prediction model.

[0042] In this embodiment, VMD assumes that the signal can be decomposed into K IMFs with different frequencies to decompose the original transformer surface vibration signal, as shown in the following formula: Where f(t) represents the original signal, t represents the current time step or index of the iteration, and u k (t) represents the Kth IMF component, A k (t)φ k (t) represents the instantaneous amplitude and phase of the IMF component.

[0043] As an implementation method, Figure 3 and Figure 4 As shown, K is 5, 6, and 7 respectively, that is, u k (t) is the 5th, 6th, and 7th components; when K=5 or K=6, the center frequencies of the components do not have frequency aliasing. However, when K=7, the components have aliasing, so K is set to 6 here.

[0044] Furthermore, considering that data loss may occur during the collection and transportation of transformer surface vibration signals, and the extracted winding signals and current signals are not of equal length, it is difficult to calculate the similarity measure between the winding extracted signals and the current signals. Therefore, the DTW algorithm is used to calculate the signal modal components; where: The main calculations are the current winding correlation R and the voltage winding interference D. The calculation formula is as follows: R i =DTW(IMF i ,I),i∈[1,6] D i =DTW(IMF i ,U),i∈[1,6] Among them, R i represents the current winding correlation of the i-th IMF component, D i Represents the voltage winding interference degree of the i-th IMF component; the sum of the first three IMFs of the intersection of R and D is taken as the transformer winding vibration signal W, and the calculation formula is as follows: In the formula, IMF i is the original vibration signal feature, a total of 6, I is the load current, U is the operating voltage; Therefore, based on the transformer vibration mechanism, the correlation between the fundamental frequency amplitude of each IMF and the current is calculated. The DTW distance between the vibration and the current is arranged in ascending order as the current winding correlation, and the DTW distance between the vibration and the voltage is arranged in descending order as the voltage winding interference. The sum of the first three IMFs is taken as the winding vibration signal. The calculation results are as follows: Figure 5 , Figure 6 As shown in Table 1; Table 1 Current winding correlation and voltage winding interference Current winding correlation Voltage winding interference [1、3、4、2、6、5] [1、4、3、2、6、5] From Table 2, it can be seen that the sum of IMF1, IMF3, and IMF4 of the decomposed signals is taken as the transformer winding vibration signal.

[0045] Furthermore, the winding vibration characteristic vector is constructed; Firstly, the transformer winding vibration signal is used to calculate the winding characteristic components from the three perspectives of time domain, frequency domain and energy domain to form the winding characteristic component data set X', namely kurtosis, average amplitude, maximum value, minimum value, skewness, vibration entropy, root mean square and total harmonic amplitude. Among them, kurtosis, average amplitude, maximum value, minimum value and skewness are selected in the time domain; vibration entropy and total harmonic amplitude are selected in the frequency domain; root mean square is selected in the energy domain to describe the operating status of the transformer winding under different working conditions; the trend of each winding characteristic component is as follows Figure 7 and Figure 8 As shown; Secondly, the historical operation data of the transformer is sorted out, and the load current, active power, and oil temperature data of the transformer operating data at the same time are taken as one vibration record, a total of 2000 records, forming an operation data set, recorded as Y'. X' and Y' are as follows: In the formula, X 1i , X 2i , X 3i , X 4i , X 5i , X 6i , X 7i Respectively represent the kurtosis, average amplitude, maximum value, minimum value, skewness, vibration entropy, and root mean square of transformer winding vibration characteristics IMF1, IMF3, and IMF4; Y 1j , Y 2j , Y 3j They represent the load current value, active power value, and oil temperature value respectively, and N is the total number of records. Thus, the overall sample data set can be obtained, and the data set Z is obtained by normalizing it through z_score, as shown below: Z = [X'; Y'] In the formula, X' is the winding vibration data, and Y' is the transformer operating condition data. The data set Z can be divided into training set, test set and validation set in proportion according to the business needs of prediction model training.

[0046] Specifically, S3 includes the following sub-steps: Based on the transformer winding vibration generation mechanism, the model parameters of the Informer prediction algorithm are determined through the relationship between the transformer winding vibration fundamental frequency and the load current; According to the model parameters, the winding vibration feature vector is used as the input of the Informer prediction model for model training to obtain the winding vibration feature prediction model.

[0047] S4. Predicting the winding characteristic vector of the transformer to be tested based on the trained winding vibration characteristic prediction model, and calculating the abnormality of the winding characteristic vector of the transformer to be tested according to the predicted value.

[0048] Specifically, S4 includes the following sub-steps: The trained winding vibration feature prediction model is used to predict the winding vibration feature vector of the transformer to be tested, and the predicted value is obtained. The winding vibration abnormality of the transformer to be tested is calculated based on the prediction error between the predicted value and the true value.

[0049] S5. Compare the abnormality with the abnormality detection threshold to determine the abnormal state of the transformer winding vibration to be tested.

[0050] Specifically, S5 includes the following sub-steps: The maximum abnormality value of the vibration characteristic vector of the normal transformer winding is used as the abnormality detection threshold; The abnormality of the vibration characteristic vector of the transformer winding to be tested is compared with the abnormality detection threshold to obtain the deviation degree of the vibration characteristic vector of the winding to be tested; The vibration state of the transformer winding to be tested is judged by the degree of deviation. If the degree of deviation does not exceed the standard degree threshold, the vibration state of the transformer winding to be tested is normal; if the degree of deviation exceeds the standard degree threshold, the vibration state of the transformer winding to be tested is abnormal.

[0051] Specifically, the method further includes: if the degree of deviation exceeds the standard degree threshold, classifying the abnormal state of transformer winding vibration into abnormal levels, specifically including: The abnormal level is divided according to the frequency and degree of exceeding the limit within a certain period of time.

[0052] As an implementation method, Fig. 9 As shown, based on the analysis of the transformer winding vibration generation mechanism, the Informer prediction algorithm is used to select the transformer operating condition factors current, active power, oil temperature and winding vibration feature vector as the input of the Informer prediction model, and the model parameters of the Informer prediction algorithm are set to train the winding vibration feature prediction model; For example, the model parameters of the Informer prediction algorithm are set to: epoch = 128, features = 11, enc_in = 12, dec_in = 12, d_model = 512, n_heads = 4, e_layers = 1, d_layers = 1, d_ff = 1024; Then, the trained Informer model is used to predict the vibration feature vector of the transformer winding to be tested, and the abnormality of the vibration feature vector of the winding to be tested is calculated using the prediction error. The maximum abnormality of the normal winding vibration feature vector is used as the abnormality detection threshold; the calculation formula of the abnormality S is as follows: In the formula, f1 and f2 are the predicted value and true value of the winding vibration characteristics in the time domain, frequency domain, and energy domain, respectively; according to the characteristics of the Gaussian probability distribution model, 0≤S≤1; the greater the error between the predicted value and the true value, the greater the abnormality S, and the greater the possibility of abnormal vibration of the transformer winding to be tested.

[0053] Furthermore, under ideal conditions, the prediction error distribution range of normal samples of transformer winding vibration is consistent with the normal samples used for training. By independently taking the maximum value of the standard deviation (variance) of the prediction error, the upper limit of the abnormality can be determined, that is, the maximum abnormality value of the normal sample is used as the reference range. This value is used as the abnormality detection threshold. For the transformer sample to be tested, it can be judged whether it is abnormal by comparing its degree of deviation from the reference range. If the degree of deviation of the transformer sample to be tested exceeds a preset threshold, the abnormality level is divided according to the frequency and degree of exceeding the limit within a certain period of time. The judgment rules are shown in Table 2. Table 2 Transformer winding abnormality determination table In this embodiment, the fluctuation of the predicted characteristic value of the winding vibration can reflect the fluctuation of the winding vibration, and further reflect the operation and state of the corresponding winding vibration. The abnormality degree is an indicator used to evaluate the degree of difference between the sample to be tested and the normal sample. It is the difference between the vibration characteristics in the characteristic vector and the characteristic vector predicted by the informer. By calculating the difference between the predicted value of the winding vibration characteristic vector and the true value of its characteristic vector, the deviation degree of the vibration data can be quantified by the abnormality, thereby identifying the possibility of abnormal vibration. The larger the value of the abnormality degree, the greater the deviation between the predicted value and the true value, and the greater the probability of abnormal winding vibration state.

[0054] It can be understood that the winding vibration is caused by the electromagnetic force F(t), and the fundamental frequency of the winding vibration is twice the fundamental frequency of the load current. The electromagnetic force F(t) is proportional to the square of the current, so the amplitude of the winding vibration is proportional to the square of the load current; the main cause of the core vibration is magnetostriction, and the fundamental frequency of the core vibration is twice the current frequency. The amplitude of the winding vibration is proportional to the square of the operating voltage. The transformer winding model is as follows: Fig.10 As shown in Figure 2, therefore, considering integrating different working condition information to predict transformer winding vibration feature vectors, the informer model is used for prediction. The principle of the informer prediction algorithm is to perform time series analysis (such as Fig.11The main purpose of the proposed method is to predict future data based on existing historical data, which can effectively capture the precise long-range correlation coupling between output and input, and thus improve the accuracy and reliability of transformer winding vibration anomaly prediction.

[0055] The above specific embodiments are preferred embodiments of the present invention, and are not intended to limit the specific implementation scope of the present invention. The scope of the present invention includes but is not limited to the present specific embodiments. All equivalent changes made in accordance with the shape, structure, and method of the present invention are within the protection scope of the present invention.

Claims

1. A method for detecting abnormal vibration of transformer windings in operation, characterized in that: The steps include: S1. Collecting surface vibration signals of transformer in operation based on vibration sensors; S2. performing winding vibration signal analysis and signal feature fusion on the surface vibration signal to obtain a winding vibration feature vector; S3. Based on a time series prediction algorithm, the winding vibration feature vector is used as input to train a winding vibration feature prediction model; S4, predicting the winding characteristic vector of the transformer to be tested based on the trained winding vibration characteristic prediction model, and calculating the abnormality of the winding characteristic vector of the transformer to be tested according to the predicted value; S5. Compare the abnormality with an abnormality detection threshold to determine the abnormal state of the transformer winding vibration to be tested.

2. The method for detecting abnormal vibration of transformer windings in operation according to claim 1, characterized in that: The S2 comprises the following sub-steps: S21, performing data cleaning and standardization processing on the surface vibration signal to obtain a surface vibration signal in a unified standard format; S22, decomposing the surface vibration signal into a plurality of signal modal components based on VMD; S23, calculating the signal modal components based on DTW to obtain a winding characteristic component data set; S24, extracting data features of the transformer historical operation data, combining the data set of the winding characteristic component with the data set to perform feature fusion, and obtaining the winding vibration feature vector.

3. The method for detecting abnormal vibration of transformer windings in operation according to claim 2, characterized in that: The S23 comprises the following sub-steps: Calculating the winding key metric of the signal modal component according to the DTW algorithm, extracting multiple data from the intersection of the winding key metric and summing them to obtain the transformer winding vibration signal; The winding characteristic component of the transformer winding vibration signal is calculated based on the transformer winding operation state, and a winding characteristic component data set of the winding characteristic component is established.

4. The method for detecting abnormal vibration of transformer windings in operation according to claim 2, characterized in that: The winding key metrics include at least current winding correlation and voltage winding interference; The signal modal component and the transformer load current are used as the first calculation sequence of the DTW algorithm, a first similarity metric between the signal modal component and the transformer load current is calculated, and the first similarity metric is used as the current winding correlation; The signal modal component and the transformer operating voltage are used as a second calculation sequence of the DTW algorithm, a second similarity metric between the signal modal component and the transformer operating voltage is calculated, and the second similarity metric is used as the voltage winding interference degree.

5. The method for detecting abnormal vibration of transformer windings in operation according to claim 2, characterized in that: The S24 comprises the following sub-steps: According to the historical operation data of the transformer, the operating data characteristics of the transformer at the same time are extracted to establish a transformer operation characteristic data set; the winding characteristic data set and the transformer operation characteristic data set are merged, and the merged data set is standardized to obtain a signal feature fusion vector, and the signal feature fusion vector is used as the winding vibration feature vector.

6. A method for detecting abnormal vibration of transformer windings in operation according to claim 3 or 5, characterized in that: The winding characteristic components include at least: vibration kurtosis, average amplitude, winding maximum value, winding minimum value, winding skewness, vibration entropy, root mean square and total harmonic amplitude; The transformer operating condition data characteristics at least include: transformer operating condition factor current characteristics, transformer power characteristics and transformer oil temperature.

7. The method for detecting abnormal vibration of transformer windings in operation according to claim 1, characterized in that: The S3 comprises the following sub-steps: Based on the transformer winding vibration generation mechanism, the model parameters of the Informer prediction algorithm are determined through the relationship between the transformer winding vibration fundamental frequency and the load current; According to the model parameters, the winding vibration feature vector is used as the input of the Informer prediction model for model training to obtain a winding vibration feature prediction model.

8. The method for detecting abnormal vibration of transformer windings in operation according to claim 1, characterized in that: The S4 comprises the following sub-steps: The trained winding vibration feature prediction model is used to predict the winding vibration feature vector of the transformer to be tested, and a predicted value is obtained. The winding vibration abnormality of the transformer to be tested is calculated based on the prediction error between the predicted value and the true value.

9. The method for detecting abnormal vibration of transformer windings in operation according to claim 8, characterized in that: The S5 comprises the following sub-steps: The maximum abnormality value of the vibration characteristic vector of the normal transformer winding is used as the abnormality detection threshold; Comparing the abnormality of the vibration characteristic vector of the transformer winding to be tested with the abnormality detection threshold to obtain the degree of deviation of the vibration characteristic vector of the transformer winding to be tested; The vibration state of the winding of the transformer to be tested is judged by the deviation degree, and if the deviation degree does not exceed the standard degree threshold, the vibration state of the transformer winding to be tested is normal; If the degree of deviation exceeds the standard degree threshold, the vibration state of the transformer winding to be tested is abnormal.

10. The method for detecting abnormal vibration of transformer windings in operation according to claim 9, characterized in that: The method further includes: if the deviation degree exceeds the standard degree threshold, classifying the abnormal state of transformer winding vibration into abnormal levels, specifically including: The abnormal level is divided according to the frequency and degree of exceeding the limit within a certain period of time.

Images