Transformer early turn-to-turn short circuit fault diagnosis method and device based on magnetic flux leakage analysis

Abstract

The embodiment of the invention discloses a method and a device for diagnosing an early turn-to-turn short circuit fault of a transformer based on magnetic flux leakage analysis. The method comprises the following steps: acquiring a magnetic flux leakage signal measured by a magnetic field sensor; carrying out preprocessing and feature extraction on the magnetic flux leakage signal; the characteristic quantity is input into a preset transformer turn-to-turn short circuit fault diagnosis model to output a fault diagnosis result, and the preset transformer turn-to-turn short circuit fault diagnosis model is obtained based on CFDM-Net training; if the fault diagnosis result is that the transformer has the turn-to-turn short circuit fault, determining the position of the transformer turn-to-turn short circuit fault according to the position of the magnetic field sensor, and determining the degree of the transformer turn-to-turn short circuit fault according to the characteristic quantity; and sending a prompt message according to the position and degree of the transformer turn-to-turn short circuit fault. According to the method, fault diagnosis is carried out by monitoring the magnetic field change in the operation process of the transformer, the transformer does not need to be disassembled or damaged, and the method has the advantage of non-invasion.

Description

Technical Field

[0001] The present application relates to the technical field of transformers, and in particular to a method and device for diagnosing early turn-to-turn short-circuit faults of transformers based on magnetic flux leakage analysis. Background Art

[0002] The transformer is one of the most important equipment in the power system, and its operating status directly affects the stability and safety of the power system. Turn-to-turn short circuit is one of the common fault types of transformers. Due to damage or aging of the inter-turn insulation, short circuits occur between adjacent coils, which in turn lead to serious consequences such as increased transformer temperature, decreased insulation performance, and even damage. Therefore, it is of great significance to diagnose transformer turn-to-turn short circuit faults in a timely and accurate manner.

[0003] Traditional transformer turn-to-turn short-circuit fault diagnosis methods mainly include electrical quantity monitoring, vibration analysis, oil chromatography analysis, etc. However, these methods have many limitations, such as the electrical quantity monitoring method is easily affected by external interference, the vibration analysis method has high requirements for fault feature extraction, and the oil chromatography analysis method requires regular sampling and a long analysis cycle. Therefore, a more accurate, real-time, and non-invasive transformer turn-to-turn short-circuit fault diagnosis method is urgently needed. Summary of the invention

[0004] The present application provides a transformer early turn-to-turn short-circuit fault diagnosis method and device based on magnetic flux leakage analysis. The method performs fault diagnosis by monitoring the magnetic field changes during the operation of the transformer, without disassembling or destroying the transformer, and has the advantage of being non-invasive; the method can also realize real-time detection of magnetic field changes during the operation of the transformer and quickly perform fault diagnosis, and has the advantage of strong real-time performance; the fault diagnosis is performed by combining signal processing technology and machine learning algorithms, and has high accuracy; in addition, the method is suitable for turn-to-turn short-circuit fault triage of transformers of various types and sizes, and has wide applicability.

[0005] In a first aspect, an embodiment of the present application provides a transformer early turn-to-turn short circuit fault diagnosis method based on magnetic leakage analysis, wherein a plurality of magnetic field sensors are arranged around the transformer; the method comprises: During the operation of the transformer, obtaining a leakage magnetic signal measured by the magnetic field sensor; Preprocessing and feature extraction are performed on the magnetic flux leakage signal to obtain a feature quantity reflecting the transformer turn-to-turn short circuit fault; Inputting the characteristic quantity into a preset transformer turn-to-turn short-circuit fault diagnosis model to output a fault diagnosis result, wherein the preset transformer turn-to-turn short-circuit fault diagnosis model is obtained based on CFDM-Net training; If the fault diagnosis result is that the transformer has a turn-to-turn short-circuit fault, determining the location of the transformer turn-to-turn short-circuit fault according to the location of the magnetic field sensor, and determining the extent of the transformer turn-to-turn short-circuit fault according to the characteristic amount; A prompt message is sent according to the location and extent of the transformer inter-turn short circuit fault.

[0006] In some embodiments, the characteristic quantity includes the complex spectrum amplitude of the leakage magnetic signal at different frequencies; and the step of determining the degree of the transformer inter-turn short circuit fault according to the characteristic quantity includes: According to the characteristic quantity, determining a first energy density and value of a high frequency band interval and a second energy density and value of a low frequency band interval; wherein the high frequency band interval includes a frequency not greater than a first preset frequency and not less than a demarcation frequency; and the low frequency band interval includes a frequency not greater than the demarcation frequency and not less than a second preset frequency; Calculate the energy ratio of the high frequency band interval to the low frequency band interval according to the first energy density sum value and the second energy density sum value; The degree of the transformer turn-to-turn short-circuit fault is determined according to the energy ratio.

[0007] In some embodiments, the step of determining a first energy density and value of a high frequency band interval and a second energy density and value of a low frequency band interval according to the characteristic quantity comprises: For the complex spectrum amplitude corresponding to the target frequency in the magnetic leakage signal in the characteristic amount, calculate the energy density corresponding to the target frequency; wherein the target frequency is a frequency not less than the second preset frequency and not greater than the first preset frequency; Adding energy densities corresponding to target frequencies that are not greater than a first preset frequency and not less than a boundary frequency to obtain the first energy density sum value; The energy densities corresponding to the target frequencies that are not greater than the boundary frequency and not less than the second preset frequency are summed to obtain the second energy density sum value.

[0008] In some embodiments, the step of calculating the energy density corresponding to the target frequency from the complex spectrum amplitude corresponding to the target frequency in the leakage magnetic signal in the characteristic quantity includes: calculating the square value of the complex spectrum amplitude to obtain the energy density corresponding to the target frequency.

[0009] In some embodiments, the step of determining the extent of the transformer turn-to-turn short circuit fault according to the energy ratio comprises: If the energy ratio is greater than a first preset energy ratio and less than or equal to a second preset energy ratio, it is determined that the degree of the transformer turn-to-turn short-circuit fault is in a normal state; If the energy ratio is greater than the second preset energy ratio and less than or equal to the third preset energy ratio, it is determined that the degree of the transformer turn-to-turn short circuit fault is a slight short circuit; If the energy ratio is greater than the third preset energy ratio and less than or equal to the fourth preset energy ratio, it is determined that the degree of the transformer turn-to-turn short circuit fault is a moderate short circuit; If the energy ratio is greater than the fourth preset energy ratio, it is determined that the degree of the transformer inter-turn short circuit fault is a severe short circuit.

[0010] In some embodiments, the step of preprocessing and feature extracting the magnetic leakage signal to obtain a feature quantity reflecting the transformer turn-to-turn short circuit fault includes: De-noising and filtering the magnetic flux leakage signal to obtain a pre-processed magnetic flux leakage signal; The pre-processed magnetic flux leakage signal is denoised by wavelet threshold to obtain the characteristic value.

[0011] In some embodiments, the method further comprises: Using several sample data, the cross-conditional fault disentanglement matching network is trained to obtain the preset transformer turn-to-turn short-circuit fault diagnosis model.

[0012] In a second aspect, an embodiment of the present application provides a transformer early turn-to-turn short circuit fault diagnosis device based on magnetic flux leakage analysis, wherein a plurality of magnetic field sensors are arranged around the transformer; the device comprises: An acquisition unit, used to acquire a leakage magnetic signal measured by the magnetic field sensor during the operation of the transformer; A processing unit, used for preprocessing and feature extraction of the magnetic flux leakage signal to obtain a feature quantity reflecting the transformer turn-to-turn short circuit fault; An input unit, used for inputting the characteristic quantity into a preset transformer turn-to-turn short-circuit fault diagnosis model to output a fault diagnosis result, wherein the preset transformer turn-to-turn short-circuit fault diagnosis model is obtained based on CFDM-Net training; a determination unit, configured to determine the position of the transformer interturn short-circuit fault according to the position of the magnetic field sensor, and determine the degree of the transformer interturn short-circuit fault according to the characteristic amount if the fault diagnosis result is that the transformer has an interturn short-circuit fault; The prompt unit is used to send a prompt message according to the location and degree of the transformer inter-turn short-circuit fault.

[0013] In a third aspect, an embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for diagnosing early transformer inter-turn short-circuit faults based on leakage magnetic analysis are implemented.

[0014] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the transformer early turn-to-turn short-circuit fault diagnosis method based on leakage magnetic analysis are implemented.

[0015] In the above embodiment, a method and device for diagnosing early turn-to-turn short-circuit faults of transformers based on leakage magnetic analysis are provided. The method performs fault diagnosis by monitoring the changes in the magnetic field during the operation of the transformer, without disassembling or destroying the transformer, and has the advantage of being non-invasive; the method can also realize real-time detection of magnetic field changes during the operation of the transformer and quickly perform fault diagnosis, and has the advantage of strong real-time performance; the fault diagnosis is performed by combining signal processing technology and machine learning algorithms, and has high accuracy; in addition, the method is suitable for turn-to-turn short-circuit fault triage of transformers of various types and sizes, and has wide applicability. The method includes: during the operation of the transformer, obtaining a leakage magnetic signal measured by the magnetic field sensor; preprocessing and feature extraction of the leakage magnetic signal to obtain a feature quantity reflecting the transformer turn-to-turn short-circuit fault; inputting the feature quantity into a preset transformer turn-to-turn short-circuit fault diagnosis model to output a fault diagnosis result, wherein the preset transformer turn-to-turn short-circuit fault diagnosis model is obtained based on CFDM-Net training; if the fault diagnosis result is that the transformer has a turn-to-turn short-circuit fault, determining the location of the transformer turn-to-turn short-circuit fault according to the location of the magnetic field sensor, and determining the degree of the transformer turn-to-turn short-circuit fault according to the feature quantity; and sending a prompt message according to the location and degree of the transformer turn-to-turn short-circuit fault. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flowchart of a transformer early turn-to-turn short circuit fault diagnosis method based on magnetic leakage analysis according to some embodiments is exemplarily shown; Figure 2 A schematic structural diagram of a transformer early turn-to-turn short circuit fault diagnosis device based on magnetic leakage analysis according to some embodiments is exemplarily shown. DETAILED DESCRIPTION

[0017] In order to make the purpose and implementation method of the present application clearer, the exemplary implementation method of the present application will be clearly and completely described below in conjunction with the drawings in the exemplary embodiments of the present application. Obviously, the described exemplary embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0018] It should be noted that the brief description of terms in this application is only for the convenience of understanding the embodiments described below, and is not intended to limit the embodiments of this application. Unless otherwise specified, these terms should be understood according to their common and usual meanings.

[0019] The terms "first", "second", "third", etc. in the specification and claims of this application and the above drawings are used to distinguish similar or similar objects or entities, and do not necessarily mean to limit a specific order or sequence, unless otherwise noted. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances.

[0020] The terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device comprising a list of components is not necessarily limited to all the components expressly listed but may include other components not expressly listed or inherent to such product or device.

[0021] Traditional transformer turn-to-turn short-circuit fault diagnosis methods mainly include electrical quantity monitoring, vibration analysis, oil chromatography analysis, etc. However, these methods have many limitations, such as the electrical quantity monitoring method is easily affected by external interference, the vibration analysis method has high requirements for fault feature extraction, and the oil chromatography analysis method requires regular sampling and a long analysis cycle. Therefore, a more accurate, real-time, and non-invasive transformer turn-to-turn short-circuit fault diagnosis method is urgently needed.

[0022] In order to solve the above technical problems, the embodiments of the present application provide a transformer early turn-to-turn short-circuit fault diagnosis method and device based on leakage magnetic analysis. The method performs fault diagnosis by monitoring the changes in the magnetic field during the operation of the transformer, without the need to disassemble or destroy the transformer, and has the advantage of non-invasiveness; the method can also realize real-time detection of magnetic field changes during the operation of the transformer and quickly perform fault diagnosis, and has the advantage of strong real-time performance; the fault diagnosis is performed by combining signal processing technology and machine learning algorithms, and has high accuracy; in addition, the method is suitable for turn-to-turn short-circuit fault triage of transformers of various types and sizes, and has wide applicability.

[0023] Figure 1 A flow chart of a transformer early turn-to-turn short circuit fault diagnosis method based on magnetic leakage analysis according to some embodiments is exemplarily shown. The method includes S100-S500.

[0024] S100: During operation of the transformer, a leakage magnetic signal measured by the magnetic field sensor is obtained.

[0025] In the embodiment of the present application, step S100 can be completed using a magnetic field detection module.

[0026] In the embodiment of the present application, multiple magnetic field sensors may be arranged around the transformer, specifically in a uniform arrangement manner, where multiple magnetic field sensors are evenly arranged around the transformer, so that the magnetic field changes during the operation of the transformer can be fully detected.

[0027] Exemplarily, 18 magnetic field sensors may be evenly arranged on the transformer. Specifically, nine magnetic field sensor measurement points are evenly selected on both sides of the primary winding and the secondary winding of the transformer, a magnetic field sensor is set at each magnetic field sensor measurement point, and the magnetic field sensor is used to obtain the leakage magnetic signal.

[0028] In some embodiments, the magnetic field sensor may be a Hall sensor, which has high sensitivity and accuracy.

[0029] S200: Preprocessing and feature extraction are performed on the magnetic flux leakage signal to obtain a feature quantity reflecting the transformer turn-to-turn short circuit fault.

[0030] In the embodiment of the present application, step S200 can be completed using a signal processing module.

[0031] In some embodiments, the step of preprocessing and feature extracting the leakage magnetic signal to obtain a characteristic quantity reflecting the transformer inter-turn short circuit fault includes: denoising and filtering the leakage magnetic signal to obtain a preprocessed leakage magnetic signal; denoising the preprocessed leakage magnetic signal through a wavelet threshold to obtain the characteristic quantity.

[0032] In this embodiment, the step of preprocessing the magnetic flux leakage signal includes steps such as denoising and filtering. The preprocessed magnetic flux leakage signal can improve the signal-to-noise ratio and reliability of the signal. Then, feature extraction is performed on the preprocessed magnetic flux leakage signal to obtain a feature quantity reflecting the transformer inter-turn short circuit fault. Specifically, the feature extraction step can be completed by a wavelet threshold denoising method.

[0033] The following introduces the basic principle of wavelet threshold denoising. The basic principle of wavelet threshold denoising is based on the fact that useful signals and noise signals have different wavelet coefficients after decomposition. By selecting appropriate thresholds to perform threshold quantization on the wavelet coefficients, useful signals can be extracted. The specific process is analyzed as follows: 1) Wavelet decomposition of the signal. Use wavelet transform to perform D-layer decomposition and obtain the wavelet coefficients corresponding to different decomposition layers of the signal .

[0034] ; In the formula, is the threshold value; is the wavelet coefficient; is the estimated wavelet coefficient; To decompose the scale; is the order of the modal components.

[0035] 2) Threshold quantization of wavelet decomposition coefficients. Select and determine an appropriate threshold, perform threshold quantization on the high-frequency wavelet coefficients of each layer of signal decomposition, remove the wavelet coefficients of different layers of noise signals, retain the wavelet coefficients of enhanced useful signals, and then obtain the wavelet coefficients .

[0036] A fixed threshold is used, which is defined as follows: ; ; In the formula, is the number of sampling points; is the high frequency coefficient of the first layer; is the estimated noise standard deviation.

[0037] 3) Wavelet reconstruction of the signal. The high-frequency wavelet coefficients of each layer of the signal after threshold processing and the low-frequency scale coefficients of the Dth layer are reconstructed using the inverse wavelet transform to obtain the denoised signal. , are the amplitude and phase parameters of the leakage magnetic signal, that is, is the complex spectrum amplitude of the magnetic flux leakage signal at different frequencies.

[0038] The above is an introduction to the basic principles of wavelet threshold denoising.

[0039] In the embodiment of the present application, in addition to the feature quantity obtained by wavelet threshold denoising, the feature quantity may also include sample entropy obtained by processing the magnetic leakage signal, and the sample entropy is used to quantify the information complexity in the magnetic leakage signal.

[0040] Specifically, it is completed through the following process: assumed is the leakage magnetic signal measured when the transformer winding fails in the early stage. Decompose into The center frequency is Discrete modal components. The constrained variational model is constructed as follows: ; In the formula, For the modal components; is the order of the modal components; For the The center frequency of the modal components; is the total number of modal components; Relative to the time series The gradient of is a time series; is the unit pulse function; It is a mathematical symbol; For the modal components about the time series Function of st It is the abbreviation of the mathematical constraint condition subject to.

[0041] Introducing a quadratic penalty coefficient and the Lagrange multiplier , transforming the constrained problem into an unconstrained variational problem. The extended Lagrangian formula is expressed as follows: ; in, is the extended Lagrangian formula, For the modal components; For the The center frequency of the modal components; is the Lagrange multiplier; is the quadratic penalty coefficient; represents the order of the modal components; Relative to the time series The gradient of is the unit pulse function; It is a mathematical symbol; is a time series; For the modal components about the time series Function of It is the leakage magnetic signal measured when the transformer winding fails in the early stage; is the Lagrange multiplier in the time domain; is the normalized proportional coefficient, which is used to adjust the energy balance of signal decomposition.

[0042] The alternating direction multiplier method is used to solve the saddle point of the above equation, thereby obtaining the solution of the constrained variational model of the original minimization problem, as follows: ; ; In the formula, For the The modal component is in the Frequency domain representation after iterations; is the number of iterations; is the order of the modal components; is the frequency; is the signal spectrum; is the index variable of the modal component; For the Frequency domain representation of the modal components; is the Lagrange multiplier in the frequency domain; is the quadratic penalty coefficient; For the The center frequency of the modal components; For the The modal component is in the The center frequency after iterations; is the mode spectrum.

[0043] The updated Lagrange multiplier formula is: ; in, for The frequency domain Lagrange multiplier after iterations; for The frequency domain Lagrange multiplier after iterations; is the step length; is the signal spectrum; is the order of the modal components; For the The modal component is in the Frequency domain representation after iterations.

[0044] In the iterative process of solving the variational model, the center frequency and bandwidth of each component are continuously updated until the iteration stop condition is met, that is: ; in, is the order of the modal components; is the number of iterations; For the The modal component is in the Frequency domain representation after iterations; For the The modal component is in the Frequency domain representation after iterations; The value of the iteration stopping condition.

[0045] The measured magnetic flux leakage signal is combined into a Dimensional time series , which is constructed as Dimensional vector ; in , is the window, with values ​​of 1, 2, ..., ; The value of the iteration stop condition. and The distance between them is: ; in, for and The distance between and is the local window vector extracted from the original time series; is a local index inside a vector; is the global index of the time series; is the embedding dimension, which can also be understood as the window length.

[0046] Pick The maximum value of the difference is and For a given ,statistics and The distance between Number of , 0.1 ~0.25 ( is the standard deviation of the time series).

[0047] definition: ; .

[0048] Finally, we get the sample entropy definition: ; in, is the sample entropy; For time series The dimension of is the embedding dimension, which can also be understood as the window length; 0.1 ~0.25 ( is the standard deviation of the time series); is the normalized window similar probability of ; is the local window vector The number of similarities with other windows within distance r; is the total number of possible windows after excluding itself; is the embedding dimension similar probability of ; is the similarity probability with embedding dimension m+1.

[0049] The sample entropy here represents the amount of information contained in the leakage magnetic signal measured by the magnetic field sensor. This step is to decompose and extract the leakage magnetic signal.

[0050] S300, inputting the characteristic quantity into a preset transformer turn-to-turn short-circuit fault diagnosis model to output a fault diagnosis result, wherein the preset transformer turn-to-turn short-circuit fault diagnosis model is obtained based on CFDM-Net (cross-conditional fault disentanglement matching network) training.

[0051] In the embodiment of the present application, before executing step S300, a preset transformer turn-to-turn short-circuit fault diagnosis model can also be obtained by a machine learning module. In some embodiments, the method further includes: using a plurality of sample data to train a cross-conditional fault disentanglement matching network (CFDM-Net) to obtain a preset transformer turn-to-turn short-circuit fault diagnosis model.

[0052] In this embodiment, the sample data includes sample feature quantities and fault diagnosis results corresponding to the sample feature quantities, and the fault diagnosis results corresponding to the sample feature quantities include the presence of a short-circuit fault between turns of the transformer and the absence of a short-circuit fault between turns of the transformer. The process of obtaining the sample feature quantities in this embodiment is to obtain a leakage magnetic signal using a magnetic field sensor in advance with a clear fault diagnosis result, and then preprocess and feature extract the leakage magnetic signal to obtain the sample feature quantities. The cross-conditional fault disentanglement matching network is trained using the sample feature quantities and the fault diagnosis results, and finally a preset transformer turn-to-turn short-circuit fault diagnosis model is obtained.

[0053] In some embodiments, a number of sample data are divided, part of the sample data is used as a training set, and part of the sample data is used as a test set. The cross-conditional fault disentanglement matching network is trained using the training set; then the trained cross-conditional fault disentanglement matching network is tested using the test set, and if the test result meets the requirements, the trained cross-conditional fault disentanglement matching network is determined as a preset transformer inter-turn short-circuit fault diagnosis model. If the test result does not meet the requirements, the cross-conditional fault disentanglement matching network is continuously trained using the training set until the test result meets the requirements.

[0054] The transformer inter-turn short-circuit fault diagnosis model preset in the embodiment of the present application is obtained based on training of a machine learning algorithm, and the machine learning algorithm uses a cross-conditional fault disentanglement matching network (CFDM-Net), which enables the model to have higher classification accuracy and generalization ability.

[0055] In the embodiment of the present application, the leakage magnetic signal is preprocessed and feature extracted to obtain a feature quantity reflecting the transformer turn-to-turn short-circuit fault, and a preset transformer turn-to-turn short-circuit fault diagnosis model is used to output a fault diagnosis result, which includes the presence of a transformer turn-to-turn short-circuit fault and the absence of a transformer turn-to-turn short-circuit fault. This step realizes the combination of signal processing technology and machine learning algorithm for fault diagnosis, with high accuracy. In the embodiment of the present application, step S300 can be performed by a fault diagnosis module.

[0056] The following introduces the principle of the cross-conditional fault disentanglement matching network. Specifically, the cross-conditional fault disentanglement matching network (CFDM-Net) is used to match the most relevant faults to the early fault in the fault-specific shared database through shared features, and associate it with the fault information of other working conditions. Then, the cal fault is realized through the partial fusion of the designed ABMixup adaptive early fault and historical fault latent layer features; finally, the fused latent layer finite element analysis is used to generate the network enhanced early fault.

[0057] In actual industrial early fault diagnosis, the fault signal is weak, and the information collected under multiple conditions at the same time - due to differentiated fault monitoring, there is a strong coupling mechanism between the parts. An effective data-driven FD subnet is designed to solve multi-condition faults. The proposed multi-source DO-MAIN distribution alignment strategy is adopted to reliably extract the shared features of the fault through the parameter sharing feature extractor. The proprietary and shared features are stored in the designed fault proprietary and shared database for matching early faults with their own faults under other working conditions. The design of the FD subnet is as follows: 1) Proprietary feature extraction: Proprietary features present different characteristics under different working conditions. The extraction of proprietary features belongs to the typical black box model and has no fixed mathematical model. The latent layer features under the same working condition are designed as positive faults, while the latent layer features under different working conditions are designed as negative faults. Positive and negative sample pairs are embedded in the comparison strategy to extract proprietary features, as shown below: ; In the formula, Different working conditions and Contrastive loss function; yes No. The fault characteristics of the potential layer, yes No. Fault characteristics of potential layers; is the weight, which affects the gradient weight of positive and negative samples; yes No. The fault characteristics of the potential layer. yes and Similarity features.

[0058] Loss Function The expression is as follows: ; In the formula, is the loss function, is the number of working conditions, and represents two working conditions; express and Correlations between working conditions.

[0059] 2) Shared feature extraction: The maximum mean difference (MMD) metric is used as the loss function to embed the strategy of sharing feature extractors for multi-source domain distribution comparison and learn the mapping matrix . It can be solved using kernel techniques,• represents a mapping relationship.

[0060] Shared parameter networks are an effective solution for learning shared features. The shared feature extractor can be optimized as follows: ; In the formula, and Indicates First working conditions. Represents a parameter A shared feature extractor for is the shared feature extractor parameters; , Respectively and A common feature extractor for and They represent the input signals under different working conditions respectively.

[0061] In order to ensure that the extracted shared features retain as much fault information in the original domain as possible, a loss function based on mutual information measurement is designed. The parameters of the shared feature extractor can be optimized from the data distribution level by maximizing the mutual information between the shared features and the original features.

[0062] ; In the formula represents the information coefficient, are the corresponding network parameters, is the shared feature extractor parameters; and Respectively represent First working conditions; and Respectively represent input signals under different working conditions; and Stored in a shared database.

[0063] 3) Related shared transfer exploration: Since shared features cover cross-condition fault information, the shared feature extractor has strong generalization ability. Therefore, the shared features of early faults can be extracted as follows: ; In the formula, Early fault common features of For A common feature extractor for .

[0064] Using shared features as transfer matching objects, the most relevant faults can be obtained from the fault-specific and shared databases. and . The expression is as follows: ; in, is obtained by matching the historical fault data The most relevant shared features; Early fault common features of For the working conditions, is the number of working conditions; is the index variable of the number of working conditions; is the maximum information coefficient, used to quantify and The correlation between them.

[0065] Then, through the internal mapping of the fault-specific and shared databases, the corresponding and , as shown below: ; in, For the Matching operating characteristics The mean parameter of For the Matching operating characteristics The standard deviation parameter of is the feature representation of the latent variable, through mapping The obtained variable is an intermediate variable and is used in subsequent fault diagnosis tasks; is obtained by matching the historical fault data The most relevant shared features, indicating characteristics of historical working conditions. and There is a strong similarity in feature distribution. and are identified as the most relevant fault pairs, so a proprietary feature extractor is used based on of It can be expressed as: ; in, and is the input data of mean and standard deviation, It is the input signal of early fault; For the The dedicated feature extractor parameters corresponding to each working condition.

[0066] 4) Adaptive deviation fusion of fault features: Linear interpolation of features will lead to linear interpolation of related targets. Therefore, ABMixup is proposed to achieve adaptive deviation fusion of latent layer features of early faults and the most relevant faults under other working conditions, which helps to enhance the early fault information under the current working conditions. The design of ABMixup is as follows: ; ; ; ; In the formula, and Represent the mean and standard deviation of the latent features after fusion; and is the input data of mean and standard deviation, and is the mean and standard deviation under a historical fault condition, is the loss function of the reconstruction-generation network; is the adaptive bias coefficient; For the Data scale parameters related to each working condition; express Take the probability from 1 / 2 to 1, Indicated in In the optimization of The probability is less than probability.

[0067] 5) Early failure enhancement: Early failure Can be used Enhanced, so the loss function of the reconstruction-generation network is designed as follows: ; In the formula, It is captured The encoder; and They are the generator and the discriminator respectively; Represents the fusion of latent layer features, and the expression is: , is standard Gaussian noise (usually obeys N(0,1)); For the discriminator For real data loss item; Represents the distribution from real data Input signals sampled in the For the discriminator to the real sample The output probability of (the probability of being judged as "true"); For the generator The loss term is used to optimize the realism of the generated samples. Represents the conditional distribution The latent variables collected in (possibly generated by an encoder); is the decision probability of the discriminator for the generated sample.

[0068] The contributing subnetworks of FD include: (1) In the proprietary feature extractor, the variability between multiple working conditions in the real industry is taken into account. By embedding an adaptive contrast training mechanism in the feature extractor, the strong style of the extracted proprietary features can be effectively guaranteed. (2) In the shared feature extractor, by jointly mapping the cross-conditional sample distribution to the shared distribution space, while maximizing the mutual information between the shared features and the original features, the multi-scale alignment of the cross-conditional sample distribution information is achieved, effectively alleviating the negative transfer properties brought by multi-source heterogeneity, so as to improve the accuracy of extracting shared features.

[0069] The above is the principle of the cross-conditional fault disentanglement matching network.

[0070] S400. If the fault diagnosis result is that the transformer has a turn-to-turn short circuit fault, determine the location of the transformer turn-to-turn short circuit fault according to the location of the magnetic field sensor, and determine the extent of the transformer turn-to-turn short circuit fault according to the characteristic value.

[0071] In the embodiment of the present application, step S400 may also be performed by a fault diagnosis module.

[0072] In the embodiment of the present application, after determining that the fault diagnosis result is that the transformer has a turn-to-turn short-circuit fault, the location and extent of the fault are further determined. In the embodiment of the present application, when determining that the fault diagnosis result is that the transformer has a turn-to-turn short-circuit fault, the source of the characteristic quantity of the preset transformer turn-to-turn short-circuit fault diagnosis model is input (i.e., the characteristic quantity is determined by the leakage magnetic signal measured by which magnetic field sensor), and then the location of the transformer turn-to-turn short-circuit fault is determined according to the location of the magnetic field sensor.

[0073] In some embodiments, the characteristic quantity includes the complex spectrum amplitude of the leakage magnetic signal at different frequencies; and the step of determining the degree of the transformer inter-turn short circuit fault according to the characteristic quantity includes: According to the characteristic quantity, a first energy density and value of a high-frequency band interval and a second energy density and value of a low-frequency band interval are determined; wherein the high-frequency band interval includes frequencies not greater than a first preset frequency and not less than a demarcation frequency; the low-frequency band interval includes frequencies not greater than the demarcation frequency and not less than a second preset frequency, and the second preset frequency is lower than the first preset frequency; exemplarily, the first preset frequency is 2kHz, the second preset frequency is 50Hz, and the demarcation frequency is 1kHz.

[0074] In some embodiments, the step of determining a first energy density sum value in a high frequency band interval and a second energy density sum value in a low frequency band interval based on the characteristic quantity includes: calculating the energy density corresponding to the target frequency for the complex spectrum amplitude corresponding to the target frequency in the leakage magnetic signal in the characteristic quantity; wherein the target frequency is a frequency not less than a second preset frequency and not greater than a first preset frequency; summing the energy densities corresponding to target frequencies not greater than the first preset frequency and not less than a boundary frequency to obtain the first energy density sum value; summing the energy densities corresponding to target frequencies not greater than the boundary frequency and not less than the second preset frequency to obtain the second energy density sum value.

[0075] In some embodiments, the step of calculating the energy density corresponding to the target frequency from the complex spectrum amplitude corresponding to the target frequency in the leakage magnetic signal in the characteristic quantity includes: calculating the square value of the complex spectrum amplitude to obtain the energy density corresponding to the target frequency.

[0076] An energy ratio between the high frequency band section and the low frequency band section is calculated according to the first energy density sum value and the second energy density sum value.

[0077] Specifically, the energy ratio can be calculated by the following formula: ; Among them, the FEAI frequency domain energy asymmetry index is the energy ratio, is the second preset frequency, is the first preset frequency, is the boundary frequency between the low frequency band and the high frequency band. The target frequency Energy density; , The leakage magnetic signal is at the target frequency The complex spectrum amplitude at .

[0078] The degree of the transformer turn-to-turn short-circuit fault is determined according to the energy ratio.

[0079] In some embodiments, multiple thresholds may be set to distinguish between different degrees of inter-turn short-circuit faults. Specifically, the step of determining the degree of the transformer inter-turn short-circuit fault according to the energy ratio includes: If the energy ratio is greater than the first preset energy ratio and less than or equal to the second preset energy ratio, the degree of the transformer inter-turn short-circuit fault is determined to be normal; wherein the second preset energy ratio is greater than the first preset energy ratio. Exemplarily, the first preset energy ratio is 0.8 and the second preset energy ratio is 1.2. In this embodiment, the energy ratio is between 0.8-1.2 (0.8 < energy ratio ≤ 1.2), and it is determined that the high-frequency and low-frequency energies are balanced, there are no obvious abnormal harmonics, and the degree of the transformer inter-turn short-circuit fault is normal.

[0080] If the energy ratio is greater than the second preset energy ratio and less than or equal to the third preset energy ratio, the degree of the transformer turn-to-turn short circuit fault is determined to be a slight short circuit; wherein the third preset energy ratio is greater than the second preset energy ratio. Exemplarily, the third preset energy ratio is 2.5. In this embodiment, the energy ratio is between 1.2-2.5 (1.2 < energy ratio ≤ 2.5), and it is determined that a small amount of turn-to-turn short circuit causes local magnetic flux distortion, a small increase in high-frequency energy, and the degree of the transformer turn-to-turn short circuit fault is a slight short circuit.

[0081] If the energy ratio is greater than the third preset energy ratio and less than or equal to the fourth preset energy ratio, the degree of the transformer inter-turn short circuit fault is determined to be a moderate short circuit; wherein the fourth preset energy ratio is greater than the third preset energy ratio. Exemplarily, the fourth preset energy ratio is 4.0. In this embodiment, the energy ratio is between 2.5-4.0 (2.5 < energy ratio ≤ 4.0), and it is determined that the number of short-circuited turns increases, the high-frequency harmonic energy of the magnetic field is significantly enhanced, and the degree of the transformer inter-turn short circuit fault is a moderate short circuit.

[0082] If the energy ratio is greater than the fourth preset energy ratio, the degree of the transformer turn-to-turn short circuit fault is determined to be a severe short circuit. In this embodiment, the energy ratio is greater than 4.0, and it is determined that a severe short circuit causes severe flux oscillation, a sharp increase in the proportion of high-frequency energy, and may be accompanied by local overheating or insulation breakdown, and the degree of the transformer turn-to-turn short circuit fault is a severe short circuit.

[0083] S500: Send a prompt message according to the location and degree of the transformer inter-turn short circuit fault.

[0084] In the embodiment of the present application, step S500 can be executed using an early warning and alarm module.

[0085] In an embodiment of the present application, a prompt message is sent to the operation and maintenance personnel according to the location and degree of the transformer inter-turn short-circuit fault. The specific prompt message can be a warning message or an alarm message, so that the operation and maintenance personnel can take corresponding measures to deal with it in time. Exemplarily, when the degree of the transformer inter-turn short-circuit fault is normal, a warning message can be sent to remind the operation and maintenance personnel that a transformer inter-turn short-circuit fault may occur at the fault location. When the degree of the transformer inter-turn short-circuit fault is a slight short circuit, a moderate short circuit, and a severe short circuit, an alarm message can be sent to remind the operation and maintenance personnel of the fault location and the recommended treatment measures, so that the operation and maintenance personnel can quickly locate and repair the fault.

[0086] In the above embodiment, a method and device for diagnosing early turn-to-turn short-circuit faults of transformers based on leakage magnetic analysis are provided. The method performs fault diagnosis by monitoring the changes in the magnetic field during the operation of the transformer, without disassembling or destroying the transformer, and has the advantage of being non-invasive; the method can also realize real-time detection of magnetic field changes during the operation of the transformer and quickly perform fault diagnosis, and has the advantage of strong real-time performance; the fault diagnosis is performed by combining signal processing technology and machine learning algorithms, and has high accuracy; in addition, the method is suitable for turn-to-turn short-circuit fault triage of transformers of various types and sizes, and has wide applicability.

[0087] The embodiment of the present application also provides a transformer early turn-to-turn short circuit fault diagnosis device based on magnetic leakage analysis, wherein a plurality of magnetic field sensors are arranged around the transformer; Figure 2 The schematic diagram of the structure of a transformer early turn-to-turn short circuit fault diagnosis device based on magnetic leakage analysis according to some embodiments is exemplarily shown. The device includes an acquisition unit 201, a processing unit 202, an input unit 203, a determination unit 204 and a prompt unit 205.

[0088] An acquisition unit, used to acquire a leakage magnetic signal measured by the magnetic field sensor during the operation of the transformer; A processing unit, used for preprocessing and feature extraction of the magnetic flux leakage signal to obtain a feature quantity reflecting the transformer turn-to-turn short circuit fault; An input unit, used for inputting the characteristic quantity into a preset transformer turn-to-turn short-circuit fault diagnosis model to output a fault diagnosis result, wherein the preset transformer turn-to-turn short-circuit fault diagnosis model is obtained based on CFDM-Net training; a determination unit, configured to determine the position of the transformer interturn short-circuit fault according to the position of the magnetic field sensor, and determine the degree of the transformer interturn short-circuit fault according to the characteristic amount if the fault diagnosis result is that the transformer has an interturn short-circuit fault; The prompt unit is used to send a prompt message according to the location and degree of the transformer inter-turn short-circuit fault.

[0089] An embodiment of the present application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the transformer early turn-to-turn short circuit fault diagnosis method based on leakage magnetic analysis are implemented.

[0090] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the transformer early turn-to-turn short circuit fault diagnosis method based on magnetic leakage analysis are implemented.

[0091] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform, or by hardware.

[0092] Those skilled in the art will appreciate that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the units or processes in the accompanying drawings are not necessarily necessary for implementing the present application. Those skilled in the art will appreciate that the units in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the description of the implementation scenario, or can be changed accordingly and located in one or more devices different from the present implementation scenario. The units of the above-mentioned implementation scenarios can be combined into one unit, or can be further split into multiple sub-units.

[0093] The above serial numbers of this application are only for description and do not represent the advantages and disadvantages of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of this application, but this application is not limited to them, and any changes that can be thought of by technicians in this field should fall within the scope of protection of this application.

Claims

1. A transformer early turn-to-turn short circuit fault diagnosis method based on magnetic leakage analysis, characterized in that: A plurality of magnetic field sensors are arranged around the transformer; the method comprises: During the operation of the transformer, obtaining a leakage magnetic signal measured by the magnetic field sensor; Preprocessing and feature extraction are performed on the magnetic leakage signal to obtain a feature quantity reflecting the transformer turn-to-turn short circuit fault; Inputting the characteristic quantity into a preset transformer turn-to-turn short-circuit fault diagnosis model to output a fault diagnosis result, wherein the preset transformer turn-to-turn short-circuit fault diagnosis model is obtained based on CFDM-Net training; If the fault diagnosis result is that the transformer has a turn-to-turn short-circuit fault, determining the location of the transformer turn-to-turn short-circuit fault according to the location of the magnetic field sensor, and determining the extent of the transformer turn-to-turn short-circuit fault according to the characteristic amount; A prompt message is sent according to the location and extent of the transformer inter-turn short circuit fault.

2. The method according to claim 1, characterized in that The characteristic quantity includes the complex spectrum amplitude of the leakage magnetic signal at different frequencies; the step of determining the degree of the transformer inter-turn short circuit fault according to the characteristic quantity includes: According to the characteristic quantity, determining a first energy density and value of a high frequency band interval and a second energy density and value of a low frequency band interval; wherein the high frequency band interval includes a frequency not greater than a first preset frequency and not less than a demarcation frequency; and the low frequency band interval includes a frequency not greater than the demarcation frequency and not less than a second preset frequency; Calculating an energy ratio between the high frequency band interval and the low frequency band interval according to the first energy density sum value and the second energy density sum value; The degree of the transformer turn-to-turn short-circuit fault is determined according to the energy ratio.

3. The method according to claim 2, characterized in that The step of determining a first energy density and value of a high frequency band interval and a second energy density and value of a low frequency band interval according to the feature quantity comprises: For the complex spectrum amplitude corresponding to the target frequency in the magnetic leakage signal in the characteristic amount, calculate the energy density corresponding to the target frequency; wherein the target frequency is a frequency not less than the second preset frequency and not greater than the first preset frequency; Adding energy densities corresponding to target frequencies that are not greater than a first preset frequency and not less than a boundary frequency to obtain the first energy density sum value; The energy densities corresponding to the target frequencies that are not greater than the boundary frequency and not less than the second preset frequency are summed to obtain the second energy density sum value.

4. The method according to claim 3, characterized in that The step of calculating the energy density corresponding to the target frequency for the complex spectrum amplitude corresponding to the target frequency in the leakage magnetic signal in the characteristic amount includes: calculating the square value of the complex spectrum amplitude to obtain the energy density corresponding to the target frequency.

5. The method according to claim 2, characterized in that: The step of determining the degree of the transformer inter-turn short circuit fault according to the energy ratio comprises: If the energy ratio is greater than a first preset energy ratio and less than or equal to a second preset energy ratio, it is determined that the degree of the transformer turn-to-turn short-circuit fault is in a normal state; If the energy ratio is greater than the second preset energy ratio and less than or equal to the third preset energy ratio, it is determined that the degree of the transformer turn-to-turn short circuit fault is a slight short circuit; If the energy ratio is greater than the third preset energy ratio and less than or equal to the fourth preset energy ratio, it is determined that the degree of the transformer turn-to-turn short circuit fault is a moderate short circuit; If the energy ratio is greater than the fourth preset energy ratio, it is determined that the degree of the transformer inter-turn short circuit fault is a severe short circuit.

6. The method according to claim 1, characterized in that The step of preprocessing and feature extracting the magnetic flux leakage signal to obtain a feature quantity reflecting the transformer turn-to-turn short circuit fault includes: De-noising and filtering the magnetic flux leakage signal to obtain a pre-processed magnetic flux leakage signal; The pre-processed magnetic flux leakage signal is denoised by wavelet threshold to obtain the characteristic value.

7. The method according to claim 1, characterized in that Also includes: Using several sample data, the cross-conditional fault disentanglement matching network is trained to obtain the preset transformer turn-to-turn short-circuit fault diagnosis model.

8. A transformer early turn-to-turn short-circuit fault diagnosis device based on magnetic flux leakage analysis, characterized in that: A plurality of magnetic field sensors are arranged around the transformer; the device comprises: An acquisition unit, used to acquire a leakage magnetic signal measured by the magnetic field sensor during operation of the transformer; A processing unit, used for preprocessing and feature extraction of the magnetic leakage signal to obtain a feature quantity reflecting the inter-turn short circuit fault of the transformer; An input unit, used for inputting the characteristic quantity into a preset transformer turn-to-turn short-circuit fault diagnosis model to output a fault diagnosis result, wherein the preset transformer turn-to-turn short-circuit fault diagnosis model is obtained based on CFDM-Net training; a determination unit, configured to determine the position of the transformer interturn short-circuit fault according to the position of the magnetic field sensor, and determine the degree of the transformer interturn short-circuit fault according to the characteristic amount if the fault diagnosis result is that the transformer has an interturn short-circuit fault; The prompt unit is used to send a prompt message according to the location and degree of the transformer inter-turn short-circuit fault.

Images