A method, system, device and storage medium for transformer vibration detection

Abstract

This application discloses a transformer vibration detection method, system, device, and storage medium. Firstly, it proposes a global-local cross-comparison attention network to enhance the interaction between the global image and locally highlighted areas. Secondly, it proposes a fault-normal cross-comparison attention network to establish a comparison between fault images and normal images, further determining the local features of the fault image, discovering more complementary parts for identification, and more effectively evaluating the stratification and classification of finer-grained faults. Thirdly, it proposes an offline training method for the dual cross-comparison attention network to further optimize the network and accelerate the online fault detection processing speed. Finally, it performs short-term fault prediction on images where no faults were detected, and continues detection on the data images after fault prediction. This solves the problems of low accuracy in vibration fault detection and localization in existing technologies, and the inability to provide early warnings of faults.

Description

Technical Field

[0001] This application relates to the field of power technology, and in particular to a method, system, device and storage medium for detecting transformer vibration. Background Technology

[0002] With the introduction of the three concepts of "intelligentization," "stabilization," and "safety" in power systems, automated computer rooms have been widely promoted and put into operation. As technology advances, sensors and other equipment are extensively deployed and used in automated computer rooms, greatly improving the level and accuracy of data acquisition. However, this has also led to a significant increase in the diversity and heterogeneity of data. Traditional methods for detecting and locating transformer vibration faults struggle to demonstrate satisfactory accuracy and timeliness in this context of big data.

[0003] Currently, there are three main categories of methods for detecting vibration faults in power transformers: The first category involves constructing an accurate mathematical-physical model based on an analytical model to diagnose the transformer. The second category involves collecting incomplete fault experience, establishing a qualitative mathematical model, and inferring the fault category through the model; a typical example is the manual decision tree method. The third category is data-driven intelligent classification methods for power transformer faults, such as support vector machines and convolutional neural networks.

[0004] However, the aforementioned detection methods have some problems. The first type of method is limited by the accuracy of the mathematical model and is not sensitive enough to external factors such as temperature and light, resulting in low positioning accuracy. The second type of method relies on past fault data, lacking timeliness and failing to update the fault database in real time, making it difficult to identify various new types of vibration faults in intelligent environments. The third type of method has some inherent defects, such as neural networks requiring high-quality learning samples and being prone to getting trapped in local optima; support vector machines excel in processing small sample data, but are essentially binary classifiers and are less efficient in handling multi-class problems. Summary of the Invention

[0005] This application provides a transformer vibration detection method, system, device, and storage medium to solve the technical problems of low accuracy in vibration fault detection and location, and the inability to provide early warning of faults in the prior art.

[0006] In view of the above, the first aspect of this application provides a method for detecting transformer vibration, the method comprising:

[0007] S1. Acquire the transformer's operating signals and perform preprocessing. The operating signals include: the transformer's coil vibration frequency signal, the temperature signals of the coil and the housing, and the brightness signals around the switch and the coil acquired by an infrared camera.

[0008] S2. Convert the running signal into a two-dimensional data matrix, and then normalize the two-dimensional data matrix to obtain a two-dimensional grayscale image.

[0009] S3. Input the two-dimensional grayscale image of the time period to be tested into the trained dual cross-contrast attention network, and output the first fault detection result of the transformer. The dual cross-contrast attention network is trained from the historical two-dimensional grayscale image. The dual cross-contrast attention network includes: a global-local cross-contrast attention network and a fault-normal cross-contrast attention network.

[0010] S4. Based on the first fault detection result, determine the two-dimensional grayscale image corresponding to the transformer that has not experienced a fault as the fault two-dimensional grayscale image, and input it into the trained short-term fault data prediction model to output the two-dimensional grayscale image of the prediction time period. At the same time, input it into the dual cross-contrast attention network to output the second fault detection result of the transformer.

[0011] Optionally, the training process of the global-local cross-contrast attention network specifically includes:

[0012] The two-dimensional grayscale image obtained in step S3 is regarded as a query matrix R = [r1; r2; ...; r] consisting of N query vectors. N The cumulative attention score S for the i-th image patch is calculated based on the attention display. i ;

[0013] Based on the cumulative attention score S i From R i Select the top T query vectors corresponding to the top T highest responses from the cumulative weights of CLS to construct a new query matrix R. l , indicating the most relevant local embedding;

[0014] Based on the output function, calculate the cross-interest between the selected local query and the global key-value pair;

[0015] The output function is:

[0016]

[0017] In the formula, As a scaling factor, the query matrix Q, key matrix K, and value matrix V are embedded from the same input X∈R through different linear transformations. N×D To calculate: Q = XW Q K = XW K V = XW V l and g are coefficients, and the output is a new matrix F. G This is the fault feature matrix;

[0018] According to the fault feature matrix F G The principal components of the fault features are used to determine the fault feature weights using GRA to obtain the fault weight matrix, which is then used as the output matrix of the global-local cross-contrast attention network.

[0019] Optionally, the training process of the fault-normal cross-contrast attention network specifically includes:

[0020] Construct a fault-normal cross-contrast attention network, and input the fault 2D grayscale image and the normal 2D grayscale image into the fault-normal cross-contrast attention network for training.

[0021] Based on the output function, the fault feature matrix and fault feature weight matrix are calculated by comparing the normal data matrix and the fault-containing data matrix according to the fault two-dimensional grayscale image and the normal two-dimensional grayscale image, and are used as the output matrix of the fault-normal cross-comparison attention network.

[0022] The output matrices of the fault-normal cross-contrast attention network and the global-local cross-contrast attention network are combined using a contribution rate function to obtain the fault feature matrix and the fault feature weight matrix.

[0023] Based on the fault feature weight matrix, the classified faults are stratified to obtain a fault classification stratified feature set. At the same time, the parameters of the contrast attention network are improved and iterated using the feature set trained by the offline network to obtain the dual cross-contrast attention network.

[0024] Optionally, the step of normalizing the two-dimensional matrix to obtain a two-dimensional grayscale image specifically includes:

[0025] Based on the normalization formula, the two-dimensional data matrix is ​​normalized to obtain a two-dimensional grayscale image.

[0026] The normalization formula is as follows:

[0027]

[0028] In the formula, L(i) is the two-dimensional data matrix, where i = 1, 2, ..., N × M; Max(L) represents the maximum value in the two-dimensional grayscale image; Min(L) represents the minimum value in the two-dimensional grayscale image; R(n,m) (n = 1, 2, ..., N, m = 1, 2, ..., M) represents the pixel intensity after image normalization; the circle function is the normalization function.

[0029] A second aspect of this application provides a transformer vibration detection system, the system comprising:

[0030] The acquisition unit is used to acquire and preprocess the transformer's operating signals, which include: the transformer's coil vibration frequency signal, the temperature signal of the coil and the housing, and the brightness signal around the switch and the coil acquired by an infrared camera.

[0031] The normalization unit is used to convert the running signal into a two-dimensional data matrix, and then normalize the two-dimensional data matrix to obtain a two-dimensional grayscale image.

[0032] The first detection unit is used to input the two-dimensional grayscale image of the time period to be tested into the trained dual cross-contrast attention network and output the first fault detection result of the transformer. The dual cross-contrast attention network is trained from the historical two-dimensional grayscale image. The dual cross-contrast attention network includes: a global-local cross-contrast attention network and a fault-normal cross-contrast attention network.

[0033] The second detection unit is used to determine the two-dimensional grayscale image corresponding to the transformer that has not experienced a fault as the fault two-dimensional grayscale image based on the first fault detection result, and input it into the trained short-term fault data prediction model to output the two-dimensional grayscale image of the prediction time period. At the same time, it is input into the dual cross-contrast attention network to output the second fault detection result of the transformer.

[0034] Optionally, the training process of the global-local cross-contrast attention network specifically includes:

[0035] The two-dimensional grayscale image obtained in step S3 is regarded as a query matrix R = [r1; r2; ...; r] consisting of N query vectors. N The cumulative attention score S for the i-th image patch is calculated based on the attention display. i ;

[0036] Based on the cumulative attention score S i From R i Select the top T query vectors corresponding to the top T highest responses from the cumulative weights of CLS to construct a new query matrix R. l , indicating the most relevant local embedding;

[0037] Based on the output function, calculate the cross-interest between the selected local query and the global key-value pair;

[0038] The output function is:

[0039]

[0040] In the formula, As a scaling factor, the query matrix Q, key matrix K, and value matrix V are embedded from the same input X∈R through different linear transformations. N×DTo calculate: Q = XW Q K = XW K V = XW V l and g are coefficients, and the output is a new matrix F. G This is the fault feature matrix;

[0041] According to the fault feature matrix F G The principal components of the fault features are used to determine the fault feature weights using GRA to obtain the fault weight matrix, which is then used as the output matrix of the global-local cross-contrast attention network.

[0042] Optionally, the training process of the fault-normal cross-contrast attention network specifically includes:

[0043] Construct a fault-normal cross-contrast attention network, and input the fault 2D grayscale image and the normal 2D grayscale image into the fault-normal cross-contrast attention network for training.

[0044] Based on the output function, the fault feature matrix and fault feature weight matrix are calculated by comparing the normal data matrix and the fault-containing data matrix according to the fault two-dimensional grayscale image and the normal two-dimensional grayscale image, and are used as the output matrix of the fault-normal cross-comparison attention network.

[0045] The output matrices of the fault-normal cross-contrast attention network and the global-local cross-contrast attention network are combined using a contribution rate function to obtain the fault feature matrix and the fault feature weight matrix.

[0046] Based on the fault feature weight matrix, the classified faults are stratified to obtain a fault classification stratified feature set. At the same time, the parameters of the contrast attention network are improved and iterated using the feature set trained by the offline network to obtain the dual cross-contrast attention network.

[0047] Optionally, the step of normalizing the two-dimensional matrix to obtain a two-dimensional grayscale image specifically includes:

[0048] Based on the normalization formula, the two-dimensional data matrix is ​​normalized to obtain a two-dimensional grayscale image.

[0049] The normalization formula is as follows:

[0050]

[0051] In the formula, L(i) is the two-dimensional data matrix, where i = 1, 2, ..., N × M; Max(L) represents the maximum value in the two-dimensional grayscale image; Min(L) represents the minimum value in the two-dimensional grayscale image; R(n,m) (n = 1, 2, ..., N, m = 1, 2, ..., M) represents the pixel intensity after image normalization; the circle function is the normalization function.

[0052] A third aspect of this application provides a transformer vibration detection device, the device comprising a processor and a memory:

[0053] The memory is used to store program code and transmit the program code to the processor;

[0054] The processor is configured to execute the steps of the transformer vibration detection method as described in the first aspect above, according to the instructions in the program code.

[0055] A fourth aspect of this application provides a computer-readable storage medium for storing program code for executing the transformer vibration detection method described in the first aspect above.

[0056] As can be seen from the above technical solutions, this application has the following advantages:

[0057] This application provides a transformer vibration detection method, comprising: S1, acquiring and preprocessing the transformer's operating signals, including: the transformer's coil vibration frequency signal, the temperature signals of the coil and the tank, and the brightness signals around the switch and coil acquired by an infrared camera; S2, converting the operating signals into a two-dimensional data matrix, and then normalizing the two-dimensional data matrix to obtain a two-dimensional grayscale image; S3, inputting the two-dimensional grayscale image of the time period to be tested into a trained dual cross-contrast attention network, and outputting the first fault detection result of the transformer, wherein the dual cross-contrast attention network is trained from historical two-dimensional grayscale images, and the dual cross-contrast attention network includes: a global-local cross-contrast attention network and a fault-normal cross-contrast attention network; S4, determining the two-dimensional grayscale image corresponding to the transformer that has not experienced a fault as the fault two-dimensional grayscale image based on the first fault detection result, and inputting it into a trained short-term fault data prediction model, outputting the two-dimensional grayscale image of the prediction time period, and simultaneously inputting it into the dual cross-contrast attention network to output the second fault detection result of the transformer.

[0058] 1. Compared with traditional transformer diagnosis methods that rely on mathematical models, this invention can comprehensively consider the influence of external environmental factors by establishing a fault feature set and a fault feature weight matrix. It is more sensitive to relevant environmental factors and avoids fault detection errors under extreme conditions.

[0059] 2. Compared with fault location methods based on convolutional neural networks, the present invention adopts an offline training model method, which greatly reduces the time spent on vibration fault detection and quickly achieves accurate identification and location of vibration faults. Attached Figure Description

[0060] Figure 1 This is a flowchart illustrating an embodiment of a transformer vibration detection method provided in this application.

[0061] Figure 2 This is a schematic diagram of the structure of an embodiment of a transformer vibration detection system provided in this application.

[0062] Figure 3 This is a schematic diagram illustrating the process of normalizing a transformed two-dimensional matrix and converting it into a two-dimensional grayscale image, as provided in the embodiments of this application.

[0063] Figure 4 This is a schematic diagram of the offline training model provided in the embodiments of this application;

[0064] Figure 5 This is a schematic diagram of the global-local cross-contrast attention network training process provided in the embodiments of this application;

[0065] Figure 6 This is a schematic diagram of the training process of the fault-normal cross-contrast attention network provided in the embodiments of this application;

[0066] Figure 7 This is a schematic diagram of the network structure of a short-term fault data prediction model based on a multi-layer neural network provided in an embodiment of this application. Detailed Implementation

[0067] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0068] Please see Figure 1 This application provides a transformer vibration detection method, including:

[0069] Step 101: Collect and preprocess the transformer's operating signals. The operating signals include: the transformer's coil vibration frequency signal, the temperature signals of the coil and the tank, and the brightness signals around the switch and coil acquired by infrared camera.

[0070] It should be noted that in the application, multimodal sensor devices, signal amplifiers, signal processing devices, fault detection and location devices, fault prediction devices, and early warning devices are set up.

[0071] The multimodal sensor device is placed at different locations on the transformer tank under test. Vibration sensors are placed around the transformer coils to detect the vibration frequency of the coils; temperature sensors are placed on the transformer coils and tank to detect the temperature data of key components; and infrared cameras are mainly placed around the transformer switches and coils to measure the temperature data of the core area.

[0072] The collected signals are saved in chronological order, and then the amplified signals for that time period are output as needed through a signal amplifier for further processing.

[0073] Step 102: Convert the running signal into a two-dimensional data matrix, and then normalize the two-dimensional data matrix to obtain a two-dimensional grayscale image;

[0074] The amplified signal during this time period is sampled, and N equally spaced signal sampling points of the same length are obtained on M sensor signals respectively. The vertical axis of the sampling points is frequency, temperature or brightness.

[0075] The data N of the sampling points and the order M of the sensors are used as the horizontal and vertical coordinates of the matrix, respectively. After filling the sampled data into the matrix, three N×M two-dimensional data matrices are obtained.

[0076] The transformed two-dimensional matrix is ​​normalized and converted into a two-dimensional grayscale image, such as... Figure 3 .

[0077]

[0078] Where L(i) represents the transformed two-dimensional matrix, where i = 1, 2, ..., N × M; Max(L) represents the maximum value in the two-dimensional grayscale image; Min(L) represents the minimum value in the two-dimensional grayscale image; R(n,m) (n = 1, 2, ..., N, m = 1, 2, ..., M) represents the pixel intensity after image normalization; the circle function is the normalization function, ensuring that the normalized pixel intensity is in the range of 0-255.

[0079] Step 103: Input the two-dimensional grayscale image of the time period to be tested into the trained dual cross-contrast attention network, and output the first fault detection result of the transformer. The dual cross-contrast attention network is trained from the historical two-dimensional grayscale image. The dual cross-contrast attention network includes: global-local cross-contrast attention network and fault-normal cross-contrast attention network.

[0080] It should be noted that the dual cross-contrast attention network consists of two parts: an offline training model and an online detection model, as follows;

[0081] Offline training of the model, such as Figure 4 :

[0082] The offline fault data is processed through steps 101-102 and then input into the global-local cross-contrast attention network. Simultaneously, the offline fault data and normal data are processed through steps 101-102 and then input into the fault-normal cross-contrast attention network. Through cross-training of these two networks, a more comprehensive feature matrix for fault classification and hierarchy is obtained.

[0083] Specifically, the training process of the global-local cross-contrast attention network is as follows: Figure 5 :

[0084] 1) The processed 2D grayscale image is viewed as a matrix R = [r1; r2; ...; r] consisting of N query vectors. N The cumulative attention score for the i-th image patch is calculated based on the attention display:

[0085]

[0086] in This indicates that the attention weights, renormalized to account for the remaining connections, are considered using the identity matrix E. This represents the cumulative weight of the class embedded in CLS. This represents matrix multiplication. In this way, we track information propagating from the input layer to higher layers.

[0087] 2) From R i Select the top T query vectors corresponding to the top T highest responses from the cumulative weights of CLS to construct a new query matrix R. l , indicating the most relevant local embedding.

[0088] 3) The cross-interest between the selected local query and the global key-value pair is calculated as follows:

[0089]

[0090] in It is a scaling factor. The query matrix, key matrix, and value matrix are embedded from the same input X∈R through different linear transformations. N×D To calculate: Q = XW Q K = XW K V = XW V S∈R N×N The attention weight matrix represents the new output matrix F. G This is the fault characteristic matrix.

[0091] 4) Based on the principal components of the fault characteristics and the pre-fault characteristic matrix F G The fault weight matrix is ​​obtained by using GRA to determine the fault feature weights.

[0092] Furthermore, the training process of the fault-normal cross-contrast attention network is as follows: Figure 6 :

[0093] 5) Input the faulty 2D grayscale image and the normal 2D grayscale image into the fault-normal cross-contrast attention network for training.

[0094] 6) Both graphs have been processed through steps S41-S44, and the output function has been modified as follows:

[0095]

[0096] The fault feature matrix and fault feature weight matrix are obtained by comparing the normal data matrix and the faulty data matrix.

[0097] 7) The global-local cross-contrast attention network focuses on the cross-comparison of highlighted and local regions in the fault map, while the fault-normal cross-contrast attention network focuses on the differences between the fault map and the normal map. By combining the output matrices of the two networks using a contribution rate function, the accurate fault feature matrix and fault feature weight matrix are obtained. The fault-normal cross-contrast attention network is shown in the figure below.

[0098] 8) Based on the characteristics of the faults (frequency, temperature, brightness), the detected faults are classified, and then the classified faults are further stratified according to the fault feature weight matrix to obtain a fault classification stratified feature set. The feature set trained by the offline network is used to improve the parameters of the attention network. This process is iterated until the error is less than a specified value, at which point the iteration stops to obtain the optimal dual cross-comparison attention network model.

[0099] Online detection section:

[0100] 9) Improve the parameters of the dual cross-contrast attention network using the fault feature set obtained from offline training, process the normalized two-dimensional matrix of the online data to obtain the detection results. If a fault exists, a red alarm light will flash through the early warning device, and the location of the sensor, i.e., the fault location, will be identified. If no fault is detected, proceed to the next step of fault prediction.

[0101] Step 104: Based on the first fault detection result, determine the two-dimensional grayscale image corresponding to the transformer that has not experienced a fault as the fault two-dimensional grayscale image, and input it into the trained short-term fault data prediction model to output the two-dimensional grayscale image of the prediction time period. At the same time, input it into the dual cross-contrast attention network to output the second fault detection result of the transformer.

[0102] It should be noted that, furthermore, S5 mainly includes 5 steps:

[0103] 1) Construct a network structure for a short-term fault data prediction model based on a multi-layer neural network using the fault feature matrix, such as... Figure 7 ;

[0104] 2) Divide the dataset of fault feature matrix and true value vector into training set, validation set and test set in a ratio of 6:3:1;

[0105] 3) The parameters of the fault data prediction model based on the multilayer neural network are tuned using the training set and the validation set to obtain the optimized model SPLF-MNN;

[0106] 4) Use the test set to calculate the prediction accuracy of the SPLF-MNN model to evaluate the model's performance. If the performance meets the requirements, use the SPLF-MNN model to predict relevant data for the next day or even the next week using the two-dimensional grayscale image after fault detection.

[0107] 5) The predicted data is tested in the constructed dual cross-comparison attention network. If a fault is detected, a yellow warning light flashes and the location of the faulty sensor is identified. If no fault is detected, a green safety light remains on.

[0108] The above is a transformer vibration detection method provided in the embodiments of this application. The following is a transformer vibration detection system provided in the embodiments of this application.

[0109] Please see Figure 2 This application provides a transformer vibration detection system. It includes:

[0110] The acquisition unit 201 is used to acquire and preprocess the transformer's operating signals. The operating signals include: the transformer's coil vibration frequency signal, the temperature signal of the coil and the tank, and the brightness signal around the switch and the coil acquired by the infrared camera.

[0111] The normalization unit 202 is used to convert the running signal into a two-dimensional data matrix, and then normalize the two-dimensional data matrix to obtain a two-dimensional grayscale image.

[0112] The first detection unit 203 is used to input the two-dimensional grayscale image of the time period to be tested into the trained dual cross-contrast attention network and output the first fault detection result of the transformer. The dual cross-contrast attention network is trained from the historical two-dimensional grayscale image. The dual cross-contrast attention network includes: a global-local cross-contrast attention network and a fault-normal cross-contrast attention network.

[0113] The second detection unit 204 is used to determine the two-dimensional grayscale image corresponding to the transformer that has not experienced a fault as the fault two-dimensional grayscale image based on the first fault detection result, and input it into the trained short-term fault data prediction model to output the two-dimensional grayscale image of the prediction time period. At the same time, it is input into the dual cross-contrast attention network to output the second fault detection result of the transformer.

[0114] Furthermore, this application embodiment also provides a transformer vibration detection device, characterized in that the device includes a processor and a memory:

[0115] The memory is used to store program code and transmit the program code to the processor;

[0116] The processor is used to execute the transformer vibration detection method described in the above method embodiments according to the instructions in the program code.

[0117] Furthermore, this application embodiment also provides a computer-readable storage medium for storing program code for executing the transformer vibration detection method described in the above method embodiment.

[0118] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0119] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0120] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0121] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0122] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0123] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0124] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0125] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting transformer vibration, characterized in that, include: S1. Acquire the transformer's operating signals and perform preprocessing. The operating signals include: the transformer's coil vibration frequency signal, the temperature signals of the coil and the housing, and the brightness signals around the switch and the coil acquired by an infrared camera. S2. Convert the running signal into a two-dimensional data matrix, and then normalize the two-dimensional data matrix to obtain a two-dimensional grayscale image. S3. Input the two-dimensional grayscale image of the time period to be tested into the trained dual cross-contrast attention network, and output the first fault detection result of the transformer. The dual cross-contrast attention network is trained from the historical two-dimensional grayscale image. The dual cross-contrast attention network includes: a global-local cross-contrast attention network and a fault-normal cross-contrast attention network. S4. Based on the first fault detection result, determine the two-dimensional grayscale image corresponding to the transformer that has not experienced a fault as the fault two-dimensional grayscale image, and input it into the trained short-term fault data prediction model to output the two-dimensional grayscale image of the prediction time period. At the same time, input it into the dual cross-contrast attention network to output the second fault detection result of the transformer.

2. The transformer vibration detection method according to claim 1, characterized in that, The training process of the global-local cross-contrast attention network specifically includes: The two-dimensional grayscale image obtained in step S3 is regarded as a query matrix R = [r1; r2; ...; r] consisting of N query vectors. N The cumulative attention score S for the i-th image patch is calculated based on the attention display. i ; Based on the cumulative attention score S i From R i Select the top T query vectors corresponding to the top T highest responses from the cumulative weights of CLS to construct a new query matrix R. l , indicating the most relevant local embedding; Based on the output function, calculate the cross-interest between the selected local query and the global key-value pair; The output function is: In the formula, As a scaling factor, the query matrix Q, key matrix K, and value matrix V are embedded from the same input X∈R through different linear transformations. N×D To calculate: Q = XW Q K = XW K V = XW V l and g are coefficients, and the output is a new matrix F. G This is the fault feature matrix; According to the fault feature matrix F G The principal components of the fault features are used to determine the fault feature weights using GRA to obtain the fault weight matrix, which is then used as the output matrix of the global-local cross-contrast attention network.

3. The transformer vibration detection method according to claim 2, characterized in that, The training process of the fault-normal cross-contrast attention network specifically includes: Construct a fault-normal cross-contrast attention network, and input the fault 2D grayscale image and the normal 2D grayscale image into the fault-normal cross-contrast attention network for training. Based on the output function, the fault feature matrix and fault feature weight matrix are calculated by comparing the normal data matrix and the fault-containing data matrix according to the fault two-dimensional grayscale image and the normal two-dimensional grayscale image, and are used as the output matrix of the fault-normal cross-comparison attention network. The output matrices of the fault-normal cross-contrast attention network and the global-local cross-contrast attention network are combined using a contribution rate function to obtain the fault feature matrix and the fault feature weight matrix. Based on the fault feature weight matrix, the classified faults are stratified to obtain a fault classification stratified feature set. At the same time, the parameters of the contrast attention network are improved and iterated using the feature set trained by the offline network to obtain the dual cross-contrast attention network.

4. The transformer vibration detection method according to claim 1, characterized in that, The next step, normalizing the two-dimensional data matrix to obtain a two-dimensional grayscale image, specifically includes: Based on the normalization formula, the two-dimensional data matrix is ​​normalized to obtain a two-dimensional grayscale image. The normalization formula is as follows: In the formula, L(i) is the two-dimensional data matrix, where i = 1, 2, ..., N × M; Max(L) represents the maximum value in the two-dimensional grayscale image; Min(L) represents the minimum value in the two-dimensional grayscale image; R(n,m) (n = 1, 2, ..., N, m = 1, 2, ..., M) represents the pixel intensity after image normalization; the circle function is the normalization function.

5. A transformer vibration detection system, characterized in that, include: The acquisition unit is used to acquire and preprocess the transformer's operating signals, which include: the transformer's coil vibration frequency signal, the temperature signal of the coil and the housing, and the brightness signal around the switch and the coil acquired by an infrared camera. The normalization unit is used to convert the running signal into a two-dimensional data matrix, and then normalize the two-dimensional data matrix to obtain a two-dimensional grayscale image. The first detection unit is used to input the two-dimensional grayscale image of the time period to be tested into the trained dual cross-contrast attention network and output the first fault detection result of the transformer. The dual cross-contrast attention network is trained from the historical two-dimensional grayscale image. The dual cross-contrast attention network includes: a global-local cross-contrast attention network and a fault-normal cross-contrast attention network. The second detection unit is used to determine the two-dimensional grayscale image corresponding to the transformer that has not experienced a fault as the fault two-dimensional grayscale image based on the first fault detection result, and input it into the trained short-term fault data prediction model to output the two-dimensional grayscale image of the prediction time period. At the same time, it is input into the dual cross-contrast attention network to output the second fault detection result of the transformer.

6. The transformer vibration detection system according to claim 5, characterized in that, The training process of the global-local cross-contrast attention network specifically includes: The two-dimensional grayscale image obtained in step S3 is regarded as a query matrix R = [r1; r2; ...; r] consisting of N query vectors. N The cumulative attention score S for the i-th image patch is calculated based on the attention display. i ; Based on the cumulative attention score S i From R i Select the top T query vectors corresponding to the top T highest responses from the cumulative weights of CLS to construct a new query matrix R. l , indicating the most relevant local embedding; Based on the output function, calculate the cross-interest between the selected local query and the global key-value pair; The output function is: In the formula, As a scaling factor, the query matrix Q, key matrix K, and value matrix V are embedded from the same input X∈R through different linear transformations. N×D To calculate: Q = XW Q K = XW K V = XW V l and g are coefficients, and the output is a new matrix F. G This is the fault feature matrix; According to the fault feature matrix F G The principal components of the fault features are used to determine the fault feature weights using GRA to obtain the fault weight matrix, which is then used as the output matrix of the global-local cross-contrast attention network.

7. The transformer vibration detection system according to claim 5, characterized in that, The training process of the fault-normal cross-contrast attention network specifically includes: Construct a fault-normal cross-contrast attention network, and input the fault 2D grayscale image and the normal 2D grayscale image into the fault-normal cross-contrast attention network for training. Based on the output function, the fault feature matrix and fault feature weight matrix are calculated by comparing the normal data matrix and the fault-containing data matrix according to the fault two-dimensional grayscale image and the normal two-dimensional grayscale image. These are used as the output matrix of the fault-normal cross-comparison attention network. The output matrices of the fault-normal cross-contrast attention network and the global-local cross-contrast attention network are combined using a contribution rate function to obtain the fault feature matrix and the fault feature weight matrix. Based on the fault feature weight matrix, the classified faults are stratified to obtain a fault classification stratified feature set. At the same time, the parameters of the contrast attention network are improved and iterated using the feature set trained by the offline network to obtain the dual cross-contrast attention network.

8. The transformer vibration detection system according to claim 5, characterized in that, The next step, normalizing the two-dimensional data matrix to obtain a two-dimensional grayscale image, specifically includes: Based on the normalization formula, the two-dimensional data matrix is ​​normalized to obtain a two-dimensional grayscale image. The normalization formula is as follows: In the formula, L(i) is the two-dimensional data matrix, where i = 1, 2, ..., N × M; Max(L) represents the maximum value in the two-dimensional grayscale image; Min(L) represents the minimum value in the two-dimensional grayscale image; R(n,m) (n = 1, 2, ..., N, m = 1, 2, ..., M) represents the pixel intensity after image normalization; the circle function is the normalization function.

9. A transformer vibration detection device, characterized in that, The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the transformer vibration detection method according to any one of claims 1-4 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code for executing the transformer vibration detection method according to any one of claims 1-4.

Images