Winding fault diagnosis model training method and related device and method
By building the time-frequency map of transformer failure and using deep learning models to predict faults, the problems of low sensitivity and insufficient analysis of transformer winding fault detection in the prior art are solved, and high sensitivity and intelligent fault diagnosis are achieved.
Patent Information
- Application Number
- CN202411602481.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-11
- Publication Date
- 2025-02-18
AI Technical Summary
The existing transformer winding fault detection methods have low sensitivity and are not objective and intelligent enough. The vibration analysis method and ultrasonic method have safety and external interference problems, which limit their wide application in modern power systems.
Using a combination method of graph construction module, feature extraction module and fault diagnosis module, the time-frequency map of the transformer in the fault situation is constructed through transient signals and oscillation attenuation response signals, the timing characteristic values are extracted, and the long and short-term memory network and Transformer are used. The model performs fault type prediction.
It improves the sensitivity of fault type prediction, realizes the intelligence of winding fault type prediction, avoids external interference and damage, and improves the accuracy and reliability of detection.
Smart Images
Figure CN119474871A_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to the technical field of transformer winding fault diagnosis, and in particular to a training method of a winding fault diagnosis model and related devices and methods. [Background technology]
[0002] Traditional transformer winding fault detection methods, such as short-circuit impedance method, capacitance change method and frequency response analysis method, are effective in some cases, but generally have disadvantages such as low sensitivity, and the analysis of fault data is not objective and intelligent enough. In addition, vibration analysis method and ultrasonic method also face problems such as test safety, large external interference and potential damage in practical applications. These defects limit the widespread application of these methods in modern power systems and fail to achieve intelligent diagnosis. [Summary of the invention]
[0003] In view of this, the present invention provides a training method for a winding fault diagnosis model and related devices and methods.
[0004] The specific technical scheme of the first embodiment of the present invention is: a training method for a winding fault diagnosis model, the winding fault diagnosis model includes a spectrum construction module, a feature extraction module and a fault diagnosis module; the spectrum construction module is used to construct a time-frequency spectrum of the transformer under fault conditions according to sample data input into the winding fault diagnosis model; the sample data includes transient signals of different fault types, and an oscillation attenuation response signal obtained by applying the transient signal as an excitation source to the transformer winding port; the feature extraction module is used to extract features from the time-frequency spectrum to obtain a time series feature value in the time-frequency spectrum; the fault diagnosis module is used to predict the fault type of the transformer according to the time series feature value; the training method includes: obtaining sample data, training the winding fault diagnosis model with the sample data, and obtaining fault type prediction data; obtaining the prediction accuracy of the winding fault diagnosis model according to the fault type prediction data and the fault type of the sample data; when the prediction accuracy is lower than the preset accuracy, iteratively training the winding fault diagnosis model with the sample data until the prediction accuracy is greater than or equal to the preset accuracy, and then terminating the training.
[0005] Preferably, the use of sample data to train the winding fault diagnosis model to obtain fault type prediction data includes: the spectrum construction module constructs a time-frequency spectrum of the transformer under fault conditions according to the sample data; the feature extraction module performs feature extraction on the time-frequency spectrum to obtain time series feature values in the time-frequency spectrum; and the fault diagnosis module predicts the fault type of the transformer according to the time series feature values.
[0006] Preferably, constructing the time-frequency spectrum of the transformer under fault conditions based on the sample data includes: performing time-frequency transformation on the transient signal and the oscillation attenuation response signal in the sample data to obtain a first frequency domain sampling signal of the transient signal and a second frequency domain sampling signal of the oscillation attenuation response signal; and obtaining the time-frequency spectrum based on the first frequency domain sampling signal and the second frequency domain sampling signal.
[0007] Preferably, obtaining the time-frequency spectrum according to the first frequency domain sampling signal and the second frequency domain sampling signal includes: obtaining an admittance sequence curve and an impedance sequence curve of the transformer according to the first frequency domain sampling signal and the second frequency domain sampling signal; superimposing the admittance sequence curve, the impedance sequence curve and the second frequency domain sampling signal through a multi-dimensional channel to obtain a multi-channel superimposed time-frequency spectrum.
[0008] Preferably, the admittance sequence curve and the impedance sequence curve are obtained using the following formula:
[0009]
[0010] Among them, R out (k) is the second frequency domain sampling signal, V in (k) is the first frequency domain sampling signal, Hd is the admittance sequence curve, and Hz is the impedance sequence curve.
[0011] Preferably, before the spectrum construction module constructs the time-frequency spectrum of the transformer under fault conditions according to the sample data, it also includes: preprocessing the sample data to obtain preprocessed sample data; the preprocessing includes denoising; then the spectrum construction module constructs the time-frequency spectrum of the transformer under fault conditions according to the sample data, including: the spectrum construction module constructs the time-frequency spectrum of the transformer under fault conditions according to the preprocessed sample data.
[0012] Preferably, the fault diagnosis module predicts the fault type of the transformer according to the timing characteristic values, including: presetting an encoder and a decoder in the fault diagnosis module; using the encoder to positionally encode the timing characteristic values to obtain a sequence with position information; inputting the sequence with position information into the decoder for fault prediction to obtain the fault type of the transformer.
[0013] The specific technical solution of the second embodiment of the present invention is: a winding fault diagnosis method, the method comprising: obtaining sample data to be predicted; inputting the sample data to be predicted into a winding fault diagnosis model, wherein the winding fault diagnosis model is obtained by a training method of a winding fault diagnosis model of a transformer as described in any one of the first embodiments of the present application; obtaining a target fault type of the sample to be predicted output by the winding fault diagnosis model.
[0014] The specific technical scheme of the third embodiment of the present invention is: a winding fault diagnosis model training device, the winding fault diagnosis model includes a spectrum construction module, a feature extraction module and a fault diagnosis module, the output end of the spectrum construction module is connected to the input end of the feature extraction module, and the output end of the feature extraction module is connected to the input end of the fault diagnosis module; the spectrum construction module is used to construct a time-frequency spectrum of the transformer under fault conditions according to the sample data input to the winding fault diagnosis model; the sample data includes transient signals of different fault types, and an oscillation attenuation response signal obtained by applying the transient signal as an excitation source to the transformer winding port; the feature extraction module is used to extract features from the time-frequency spectrum to obtain the time series feature value in the time-frequency spectrum; the fault diagnosis module is used to predict the fault type of the transformer according to the time series feature value; the training device includes: a data acquisition module, used to acquire training data; a training module, used to iteratively train the winding fault diagnosis model using the training data until the prediction accuracy of the winding fault diagnosis model is greater than or equal to the preset accuracy, and the training is terminated.
[0015] The specific technical solution of the fourth embodiment of the present invention is: a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform each step of the method described in any one of the first embodiments of the present application or the second embodiment of the present application.
[0016] Implementing the embodiments of the present invention will have the following beneficial effects:
[0017] The present invention constructs a time-frequency spectrum of the transformer under fault conditions based on transient signals and oscillation attenuation response signals, extracts features from the time-frequency spectrum, obtains time series characteristic values, and predicts the fault type of the transformer based on the time series characteristic values. The time-frequency spectrum constructed based on transient signals and oscillation attenuation response signals can simultaneously characterize the time series characteristics, frequency domain characteristics, and global characteristics of the transformer signal, and predicts the fault type based on the time-frequency spectrum, effectively improving the sensitivity of the fault type prediction, and is not easily affected by external interference and damage, thereby realizing the intelligent prediction of the winding fault type.
Brief Description of the Drawings
[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0019] Figure 1 A flowchart of the steps of a training method for a winding fault diagnosis model;
[0020] Figure 2 It is the structural diagram of the winding fault diagnosis model;
[0021] Figure 3a It is a schematic diagram of the time-frequency transformation result of the transient signal;
[0022] Figure 3b It is a schematic diagram of the time-frequency transformation result of the oscillation attenuation response signal;
[0023] Figure 4 It is a schematic diagram of the admittance sequence curve;
[0024] Figure 5 It is a schematic diagram of multi-channel superposition time-frequency spectrum;
[0025] Figure 6 The curve diagram of the accuracy of the winding fault diagnosis model changing with the number of iterations;
[0026] Among them, 201 is a graph construction module; 202 is a feature extraction module; 203 is a fault diagnosis module. [Specific implementation method]
[0027] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0028] The terms "first", "second", etc. in the specification, claims and drawings of this application are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or modules is not limited to the listed steps or modules, but optionally includes steps or modules that are not listed, or optionally includes other steps or modules that are inherent to these processes, methods, products or devices.
[0029] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.
[0030] This example uses a 35kV transformer as the experimental object, and its rated capacity S n =2000kV·A, system voltage U =35kV. A transient signal with a broadband spectrum is applied to the transformer as an excitation, and an oscilloscope is used to test the oscillation attenuation response signal of the transformer under the action of the transient signal.
[0031] See also Figure 1 , is a flowchart of the steps of a training method for a winding fault diagnosis model in the first embodiment of the present application, the training method comprising:
[0032] Step 101, obtaining sample data, using the sample data to train the winding fault diagnosis model, and obtaining fault type prediction data;
[0033] Step 102: Obtain the prediction accuracy of the winding fault diagnosis model according to the fault type prediction data and the fault type of the sample data;
[0034] Step 103: When the prediction accuracy is lower than the preset accuracy, the winding fault diagnosis model is iteratively trained using sample data until the prediction accuracy is greater than or equal to the preset accuracy, and the training is terminated.
[0035] Specifically, a transient signal with a broadband spectrum is applied to the transformer winding port as an excitation source, and a high-speed broadband digital oscilloscope, a high-speed multi-channel data acquisition card and other devices are used to collect the oscillation attenuation response signal waveform of the transformer winding under the action of the transient signal, as the original data source for subsequent processing and analysis. The improved time-frequency spectrum of the transformer is constructed by the transient signal with a broadband spectrum and the oscillation attenuation response signal of the transformer, that is, the multi-channel superposition time-frequency spectrum. The spectrum has actual physical meaning and can simultaneously characterize the timing characteristics, frequency domain characteristics and global characteristics of the transformer signal, and has rich fault information and high-dimensional data characterization capabilities. The processed transformer improved time-frequency spectrum data is input into the long short-term memory network LSTM layer, and the time series feature values in the spectrum data are extracted through the long short-term memory network and output in the form of vectors. The extracted feature values are then input into an algorithm model with a self-attention mechanism architecture, such as Transformer, for training and prediction, and the model parameters are optimized by repeated iterations to obtain a trained winding fault diagnosis model. Among them, the prediction accuracy can be set according to specific needs, for example, it can be 80%, which is not limited here.
[0036] In the specific embodiments, see Figure 2 The winding fault diagnosis model includes a spectrum construction module 201, a feature extraction module 202 and a fault diagnosis module 203; the spectrum construction module 201 is used to construct a time-frequency spectrum of the transformer under fault conditions according to sample data input into the winding fault diagnosis model; the sample data includes transient signals of different fault types, and an oscillation attenuation response signal obtained by applying the transient signal as an excitation source to the transformer winding port; the feature extraction module 202 is used to extract features from the time-frequency spectrum to obtain time series feature values in the time-frequency spectrum; the fault diagnosis module 203 is used to predict the fault type of the transformer according to the time series feature values.
[0037] The time-frequency spectrum of the transformer under fault conditions is constructed based on transient signals and oscillation attenuation response signals, and the time-frequency spectrum is feature extracted to obtain time series characteristic values, and the fault type of the transformer is predicted based on the time series characteristic values. The time-frequency spectrum constructed based on transient signals and oscillation attenuation response signals can simultaneously characterize the time series characteristics, frequency domain characteristics and global characteristics of the transformer signal, and predict the fault type based on the time-frequency spectrum, effectively improving the sensitivity of the fault type prediction, and is not easily affected by external interference and damage, thus realizing the intelligent prediction of the winding fault type.
[0038] In a specific embodiment, the winding fault diagnosis model is trained using sample data to obtain fault type prediction data, including: the spectrum construction module constructs a time-frequency spectrum of the transformer under fault conditions according to the sample data; the feature extraction module extracts features from the time-frequency spectrum to obtain time series feature values in the time-frequency spectrum; the fault diagnosis module predicts the fault type of the transformer according to the time series feature values.
[0039] Specifically, the time-frequency spectrum of the transformer under fault conditions is constructed based on the sample data, and the time-frequency spectrum is input into the LSTM layer of the intelligent diagnosis model. After passing through the input gate, forget gate, output gate and other links of the long short-term memory network, the time series feature value can be extracted, which can reflect the time characteristics of the entire data. Finally, the extracted time series features are obtained, and the sequence order information is maintained, and then the sequence with position information is obtained through the position encoding of the Transformer. The above sequence with position information is input into the Transformer layer for training, and each group of position data input into the encoding link will pass through an encoding layer. It passes through the multi-head attention layer, the full connection layer, the feedforward neural network, and finally an Add&Normalize to end the process; the output of the last encoding layer will be retained; using the powerful feature value extraction capability of the Transformer, the global features of the data can be further extracted to enhance the ability to distinguish complex faults. At the same time, multiple recognition training of a large amount of data can further increase the accuracy of fault diagnosis. Through repeated training, the intelligent diagnosis model continuously optimizes parameters.
[0040] In a specific embodiment, the time-frequency spectrum of the transformer under fault conditions is constructed based on sample data, including: performing time-frequency transformation on the transient signal and the oscillation attenuation response signal in the sample data to obtain a first frequency domain sampling signal of the transient signal and a second frequency domain sampling signal of the oscillation attenuation response signal; and obtaining the time-frequency spectrum based on the first frequency domain sampling signal and the second frequency domain sampling signal.
[0041] Specifically, first, the transient signal and the oscillation decay response signal are subjected to time-frequency transformation such as Fourier transform and Hilbert transform to obtain their spectrum information.
[0042] Specifically, the Fourier transform can convert the time domain signal to the frequency domain, which can better reflect the high-frequency and low-frequency signals of the signal. The formula of the Fourier transform is as follows:
[0043]
[0044] If the Hilbert transform is used, the output of the Hilbert transform is an analytical signal, the real part of which is the original signal and the imaginary part is its Hilbert transform. The formula of the Hilbert transform is as follows:
[0045]
[0046] Through the Hilbert transform, an analytical signal can be constructed from a real signal to extract the instantaneous amplitude and instantaneous phase, which is very important for the analysis of non-stationary signals. The analytical signal z(t) is obtained by combining the real signal x(t) with its Hilbert transform. Combination.
[0047]
[0048] After obtaining the analytical signal, the instantaneous amplitude A(t) and instantaneous phase φ(r) of the signal can be further extracted:
[0049]
[0050] The calculation of amplitude and phase here can better highlight the timing correlation of the signal. For specific time-frequency transformation results of transient signals and oscillation decay response signals, please refer to Figure 3a and Figure 3b shown.
[0051] In a specific embodiment, the time-frequency spectrum is obtained according to the first frequency domain sampling signal and the second frequency domain sampling signal, including: obtaining the admittance sequence curve and the impedance sequence curve of the transformer according to the first frequency domain sampling signal and the second frequency domain sampling signal; superimposing the admittance sequence curve, the impedance sequence curve and the second frequency domain sampling signal through a multi-dimensional channel to obtain a multi-channel superimposed time-frequency spectrum. Specifically, the impedance / admittance sequence curve of the transformer is constructed through the frequency domain amplitude information of the signals at both ends.
[0052] In a specific embodiment, the admittance sequence curve and the impedance sequence curve are obtained using the following formula:
[0053]
[0054] Among them, R out (k) is the second frequency domain sampling signal, V in (k) is the first frequency domain sampling signal, Hd is the admittance sequence curve, and Hz is the impedance sequence curve. Specifically, the admittance sequence curve Figure 4 shown.
[0055] Specifically, the following formula is used to obtain the admittance sequence curve and the impedance sequence curve:
[0056] V in (k) = f1(V in (n))
[0057] R out(k) = f2(R out (n))
[0058]
[0059] Where: V in (n) is the N-point time domain sampling signal of the transient signal voltage of the winding with a broadband spectrum, V in (k) is V in The time-frequency transformation result of (n) can be either the amplitude of the fast Fourier transform or the amplitude after the Hilbert transform, where f1 represents the transform function; R out (n) is the N-point time domain sampling signal of the winding oscillation attenuation response voltage / current, R out (k) is R out The time-frequency transformation result of (n) can be either the amplitude of the fast Fourier transform or the amplitude after the Hilbert transform, f2 represents the transformation function; Hd represents the admittance sequence curve of the transformer, and Hz represents the impedance sequence curve of the transformer. The admittance / impedance frequency domain sequence curve and the oscillation attenuation response time domain signal are directly superimposed together through multi-dimensional channels to form two sequences before and after, thereby constructing a multi-channel superimposed time-frequency spectrum. This spectrum has actual physical meaning and can simultaneously characterize the timing characteristics, frequency domain characteristics and global characteristics of the transformer signal. Specifically, in this example, only three faults are selected: short circuit (SC), axial displacement (AD), and radial displacement (RD). Different transient signals with broadband spectra are applied to the transformers under these three faults and in healthy conditions, and the oscillation attenuation response data of each frequency are collected. There are 54 groups of SC faults, 65 groups of AD faults, and 44 groups of RD faults. Please refer to the corresponding multi-channel superimposed time-frequency spectrum schematic diagram. Figure 5 shown.
[0060] In a specific embodiment, before the spectrum construction module constructs the time-frequency spectrum of the transformer under fault conditions according to the sample data, it also includes: preprocessing the sample data to obtain preprocessed sample data; the preprocessing includes denoising; then the spectrum construction module constructs the time-frequency spectrum of the transformer under fault conditions according to the sample data, including: the spectrum construction module constructs the time-frequency spectrum of the transformer under fault conditions according to the preprocessed sample data.
[0061] Specifically, the measured data is preprocessed to reduce interference first. Wavelet transform can be used to denoise the acquired raw data, remove noise components, and retain useful signal information. Wavelet transform is used in this example, and its mathematical expression is:
[0062]
[0063] Where: W(a,b) is the wavelet coefficient of the signal x(t) at scale a and position b. a is the scale parameter, which determines the expansion and contraction of the wavelet function. b is the translation parameter, which determines the position of the wavelet function. ψ(t) is the mother wavelet function, and x(t) is the original sample data. The wavelet transform decomposes the signal into wavelet coefficients of different scales and positions. These coefficients contain the local characteristics and frequency information of the signal. The wavelet coefficients are then thresholded, and the coefficients less than a certain threshold are set to zero (hard threshold) or softened (soft threshold) to remove the noise component. The processed wavelet coefficients are inversely transformed to obtain the denoised signal.
[0064] In a specific embodiment, the fault diagnosis module predicts the fault type of the transformer according to the time series feature value, including: presetting an encoder and a decoder in the fault diagnosis module; using the encoder to positionally encode the time series feature value to obtain a sequence with position information; inputting the sequence with position information into the decoder for fault prediction to obtain the fault type of the transformer. Specifically, the processed transformer improved time-frequency spectrum data is input into the LSTM layer of the long short-term memory network, and the time series feature values in the spectrum data are extracted through the long short-term memory network and output in the form of a vector. The extracted feature values are then input into an algorithm model with a self-attention mechanism architecture, such as Transformer, for training and prediction. The model parameters are optimized through repeated iterations to obtain a trained winding fault diagnosis model.
[0065] The present invention applies a transient signal with a broadband spectrum to the transformer winding to achieve high-sensitivity detection of faults; and uses a time series data processing model based on a long short-term memory network combined with a self-attention model mechanism architecture algorithm to achieve intelligent diagnosis of faults, so as to overcome the deficiencies in the prior art, especially to improve the sensitivity, accuracy, reliability and automation level of fault detection and diagnosis. The present invention applies a transient signal with a broadband spectrum to the transformer winding, measures the oscillation attenuation response signal of the transformer, and uses the above signal to construct an improved time-frequency spectrum of the transformer, that is, a multi-channel superposition time-frequency spectrum, which can characterize the time series characteristics, frequency domain characteristics and global characteristics of the transformer fault signal. The LSTM-Transformer model solves the problems of low accuracy, high cost and cumbersome methods of current transformer fault detection technology. The present invention automatically identifies fault features and generates diagnostic reports through a trained model, solves the defect of relying on manual judgment of faults in the past, greatly reduces the dependence on professionals, and reduces the risk of misjudgment.
[0066] In a specific embodiment, the second embodiment of the present application provides a winding fault diagnosis method, the method comprising: obtaining sample data to be predicted; inputting the sample data to be predicted into a winding fault diagnosis model, wherein the winding fault diagnosis model is obtained by a training method of a transformer winding fault diagnosis model as described in any one of the first embodiments of the present application; obtaining the target fault type of the sample to be predicted output by the winding fault diagnosis model. The transient signal with a broadband spectrum newly measured when the transformer is in other states and the oscillation attenuation response signal are processed and input into the above-mentioned winding fault diagnosis model, and the fault diagnosis result is output, thereby improving the accuracy of fault diagnosis identification.
[0067] In a specific embodiment, it is a structural schematic diagram of a winding fault diagnosis model training device in the third embodiment of the present application, the winding fault diagnosis model includes a spectrum construction module, a feature extraction module and a fault diagnosis module, the output end of the spectrum construction module is connected to the input end of the feature extraction module, and the output end of the feature extraction module is connected to the input end of the fault diagnosis module; the spectrum construction module is used to construct a time-frequency spectrum of the transformer under fault conditions according to the sample data input to the winding fault diagnosis model; the sample data includes transient signals of different fault types, and an oscillation attenuation response signal obtained by applying the transient signal as an excitation source to the transformer winding port; the feature extraction module is used to extract features from the time-frequency spectrum to obtain the time series feature value in the time-frequency spectrum; the fault diagnosis module is used to predict the fault type of the transformer according to the time series feature value; the training device includes: a data acquisition module, used to acquire training data; a training module, used to iteratively train the winding fault diagnosis model using the training data until the prediction accuracy of the winding fault diagnosis model is greater than or equal to the preset accuracy, and the training is terminated. Figure 6 The curve diagram of the accuracy of the winding fault diagnosis model as the number of iterations changes. Figure 6 It can be found that the training process is a process in which the accuracy gradually increases and the loss decreases. When the model is iterated about 100 times, the recognition accuracy is 99.8% for the training set and 98.8% for the validation set.
[0068] In a specific embodiment, the fourth embodiment of the present application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes each step of the method described in any one of the first embodiments of the present application or the second embodiment of the present application.
[0069] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
[0070] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.
Claims
1. A training method for a winding fault diagnosis model, characterized in that: The winding fault diagnosis model includes a graph construction module, a feature extraction module and a fault diagnosis module; The spectrum construction module is used to construct a time-frequency spectrum of the transformer under fault conditions according to sample data input into the winding fault diagnosis model; the sample data includes transient signals of different fault types, and an oscillation attenuation response signal obtained by applying the transient signal as an excitation source to the transformer winding port; The feature extraction module is used to extract features from the time-frequency spectrum to obtain time series feature values in the time-frequency spectrum; The fault diagnosis module is used to predict the fault type of the transformer according to the time series characteristic value; The training method comprises: Acquire sample data, use the sample data to train the winding fault diagnosis model, and obtain fault type prediction data; Obtaining the prediction accuracy of the winding fault diagnosis model according to the fault type prediction data and the fault type of the sample data; When the prediction accuracy is lower than the preset accuracy, the winding fault diagnosis model is iteratively trained using sample data until the prediction accuracy is greater than or equal to the preset accuracy, and the training is terminated.
2. The training method of the winding fault diagnosis model according to claim 1, characterized in that: The method of using sample data to train the winding fault diagnosis model to obtain fault type prediction data includes: The spectrum construction module constructs a time-frequency spectrum of the transformer under fault conditions according to the sample data; The feature extraction module extracts features from the time-frequency spectrum to obtain time series feature values in the time-frequency spectrum; The fault diagnosis module predicts the fault type of the transformer according to the time series characteristic value.
3. The training method of the winding fault diagnosis model according to claim 2, characterized in that: The step of constructing a time-frequency spectrum of the transformer under fault conditions according to the sample data includes: Performing a data time-frequency transformation on the transient signal and the oscillation attenuation response signal in the sample data to obtain a first frequency domain sampling signal of the transient signal and a second frequency domain sampling signal of the oscillation attenuation response signal; The time-frequency spectrum is obtained according to the first frequency domain sampling signal and the second frequency domain sampling signal.
4. The training method of the winding fault diagnosis model according to claim 3, characterized in that: The step of obtaining the time-frequency spectrum according to the first frequency domain sampling signal and the second frequency domain sampling signal comprises: Obtaining an admittance sequence curve and an impedance sequence curve of the transformer according to the first frequency domain sampling signal and the second frequency domain sampling signal; The admittance sequence curve, the impedance sequence curve and the second frequency domain sampling signal are superimposed through a multi-dimensional channel to obtain a multi-channel superimposed time-frequency spectrum.
5. The training method for a winding fault diagnosis model according to claim 4, wherein the admittance sequence curve and the impedance sequence curve are obtained using the following formula: in, R out (k) is the second frequency domain sampling signal, V in (k) is the first frequency domain sampling signal, Hd is the admittance sequence curve, and Hz is the impedance sequence curve.
6. The training method of the winding fault diagnosis model according to claim 2, before the spectrum construction module constructs the time-frequency spectrum of the transformer under fault conditions according to the sample data, further comprising: Preprocessing the sample data to obtain preprocessed sample data; The preprocessing includes denoising; The spectrum construction module constructs a time-frequency spectrum of the transformer under fault conditions according to the sample data, including: The spectrum construction module constructs a time-frequency spectrum of the transformer under fault conditions according to the preprocessed sample data.
7. The training method of the winding fault diagnosis model according to claim 2, characterized in that: The fault diagnosis module predicts the fault type of the transformer according to the time series characteristic value, including: The fault diagnosis module is pre-set with an encoder and a decoder; Using the encoder to perform position encoding on the time series feature value to obtain a sequence with position information; The sequence with position information is input into the decoder for fault prediction to obtain the fault type of the transformer.
8. A winding fault diagnosis method, characterized in that: The method comprises: Obtain sample data to be predicted; Inputting the sample data to be predicted into a winding fault diagnosis model, wherein the winding fault diagnosis model is obtained by a training method for a transformer winding fault diagnosis model according to any one of claims 1 to 7; The target fault type of the sample to be predicted output by the winding fault diagnosis model is obtained.
9. A winding fault diagnosis model training device, characterized in that: The winding fault diagnosis model includes a graph construction module, a feature extraction module and a fault diagnosis module, wherein the output end of the graph construction module is connected to the input end of the feature extraction module, and the output end of the feature extraction module is connected to the input end of the fault diagnosis module; The spectrum construction module is used to construct a time-frequency spectrum of the transformer under fault conditions according to sample data input into the winding fault diagnosis model; the sample data includes transient signals of different fault types, and an oscillation attenuation response signal obtained by applying the transient signal as an excitation source to the transformer winding port; The feature extraction module is used to extract features from the time-frequency spectrum to obtain time series feature values in the time-frequency spectrum; The fault diagnosis module is used to predict the fault type of the transformer according to the time series characteristic value; The training device comprises: A data acquisition module, used to acquire training data; The training module is used to iteratively train the winding fault diagnosis model using the training data until the prediction accuracy of the winding fault diagnosis model is greater than or equal to the preset accuracy, and then the training is terminated.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor is caused to perform each step of the method as claimed in any one of claims 1 to 7 or claim 8.