Method and system for detecting high-resistance grounding fault of power distribution network

Abstract

The invention discloses a power distribution network high-resistance grounding fault detection method and system, and relates to the technical field of power distribution network fault detection. A power distribution network high-resistance grounding fault detection method comprises the following steps: S1, collecting a transient zero-sequence voltage signal of a target feeder line as a to-be-decomposed electric quantity signal; s2, variational mode decomposition is carried out on the to-be-decomposed electric quantity signals, mode functions obtained by decomposing the single to-be-decomposed electric quantity signals are arranged according to the center frequency from low to high, and one-dimensional feature vectors of the to-be-decomposed electric quantity signals are obtained; and S3, inputting the one-dimensional feature vector of the electric quantity signal to be decomposed into a pre-constructed Transform model to obtain a detection result. According to the method, the high-resistance grounding fault and the disturbance condition can be accurately identified, the fault identification accuracy is very high, and the power distribution network fault detection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault detection, and in particular to a distribution network high-resistance grounding fault detection method and system. Background Art

[0002] In remote mountainous areas with low population density, medium-voltage distribution networks often have long transmission distances and complex terrain. Distribution lines or branches need to cross grasslands and forests to supply power to distant residential areas. When crossing forest grasslands, the lines are prone to overlap with trees and grasslands to form ground faults. Such fault media usually have high impedance, up to several thousand ohms to tens of kilohms, which is called high impedance fault (HIF). Since the fault characteristics of HIF are weak and the fault mutation is small, it is difficult for the commonly used protection devices in the distribution network to accurately detect and respond to such faults. However, if such faults are not discovered and cleared in time after they occur, the long-term accumulation of electrical energy is very likely to ignite vegetation and cause wildfires, which will not only cause serious damage to the ecological environment, but also threaten human life and property safety.

[0003] Therefore, timely and accurate detection and identification of high-resistance grounding faults in distribution networks is of vital importance to preventing the occurrence of wildfire disasters. Summary of the invention

[0004] The technical problem to be solved by the present invention is that the HIF fault characteristics are weak and the fault mutation amount is small under the existing background, which makes it difficult to be accurately detected and responded to by the commonly used protection devices in the distribution network. The purpose is to provide a distribution network high-resistance grounding fault detection method and system, which can quickly and effectively detect and identify such faults, improve the distribution network fault detection efficiency, and have important practical significance for preventing the occurrence of forest fire disasters.

[0005] The present invention is achieved through the following technical solutions:

[0006] A method for detecting a high-resistance grounding fault in a distribution network comprises the following steps:

[0007] S1, collecting the transient zero-sequence voltage signal of the target feeder as the electric quantity signal to be decomposed;

[0008] S2. performing variational modal decomposition on the electric quantity signal to be decomposed, wherein the modal functions obtained by decomposing a single electric quantity signal to be decomposed are arranged to obtain a one-dimensional feature vector of the electric quantity signal to be decomposed;

[0009] S3. Input the one-dimensional feature vector of the electric quantity signal to be decomposed into the pre-built Transformer model to obtain the detection result.

[0010] As a possible design, the above step S2 includes:

[0011] The quantity signal to be decomposed is preset as f, and the quantity signal to be decomposed is decomposed into K modal components μ according to variational modal decomposition. k , the center frequency of the modal classification is ω k , taking the sum of the decomposed modal functions equal to the input signal as the constraint condition, the constrained variational equation is obtained, and then the constrained variational equation is transformed into an unconstrained variational equation according to the Lagrangian operator λ and the quadratic penalty factor α. Then, the alternating direction multiplier method is used to alternately iteratively update u(k) and ω(k) to find the saddle point of the unconstrained variational equation, thereby obtaining the optimal solution of the constrained variational model.

[0012] As a possible design, the above objective function and constraint conditions are shown in equation (1) and equation (2) respectively:

[0013]

[0014] Among them, {u k}={u1,u2,…,u K} is the kth modal component after decomposition, f(t) is the original signal, || ||2 represents the 2-norm of the orientation quantity, is the partial derivative of the function at time t, {ω k}={ω1,ω2,…,ω K} is the center frequency of each modal component, δ(t) is the unit impulse function, t represents the time variable, * is the convolution operator, and j represents the imaginary unit.

[0015] As a possible design, the above-mentioned constrained variational equation (1) is transformed into an unconstrained variational equation according to the Lagrangian operator λ and the quadratic penalty factor α, as shown in the following formula:

[0016]

[0017] Among them, λ(t) is the Lagrangian operator, α is the quadratic penalty factor, < > represents the inner product operation, represents the sum of all modal components.

[0018] As a possible design, the above-mentioned alternating direction multiplier method is used to alternately iteratively update u(k) and ω(k) to find the saddle point of the unconstrained variational equation of formula (3), thereby obtaining the optimal solution of the constrained variational model of formulas (1) and (2):

[0019]

[0020] in, and They are f(t) and u k (t), λ(t), and The function obtained after Fourier transform, n represents the number of iterations, and ω represents the conversion of the signal from time domain to frequency domain.

[0021] As a possible design, the above also includes step S0, using the one-dimensional feature vector of the electric quantity signal to be decomposed as the input data of the Transformer network, dividing the input data set into a training set and a test set, using the training set to train the Transformer model, and using the test set to test the network.

[0022] As a possible design, the overall structure of the Transformer network model in the above step S3 includes an input layer, a position embedding layer, an addition layer, a self-attention layer, a fully connected layer, and an output layer.

[0023] As a possible design, the output layer uses a variational mode decomposition algorithm to adaptively decompose the transient zero-sequence voltage signal, converts the decomposition result into one-dimensional time series data, and serves as the input of the Transformer network;

[0024] The position embedding layer maps the input features to a 256-dimensional latent space through the position embedding layer, and introduces a learnable position encoding to enhance the representation of the position in the sequence;

[0025] The addition layer fuses the features of the input layer and the position embedding layer, so that the model can simultaneously utilize the content information and position information of the input sequence, thereby enhancing the expressiveness and effect of the self-attention mechanism;

[0026] The self-attention layer uses two layers of self-attention layers to calculate the attention weight corresponding to each position. The first layer uses a causal mask to ensure that only the current position and the information before it can be seen at each position. The second layer of self-attention layer further enhances the model's global understanding of the overall sequence and its ability to learn complex features. Each layer is a self-attention layer with 4 heads and 128 keys.

[0027] The fully connected layer linearly maps the high-order features extracted by the self-attention layer to the binary classification results for classification output;

[0028] The output layer classifies the disturbance conditions into one category, thus performing a binary classification task. The number of neurons is 2, and the input of the output layer is the output of the fully connected layer, which is output as a probability vector.

[0029] As a possible design, the above output layer adopts a Softmax classifier.

[0030] As a possible design, each of the above self-attention layers is a self-attention layer with 4 heads and 128 keys.

[0031] As a possible design, the disturbance conditions are classified into one category in the model of the above output layer, so the model performs a binary classification task. The number of neurons is 2, and the input of the output layer is the output of the fully connected layer, which is output as a probability vector.

[0032] A distribution network high-resistance grounding fault detection system, comprising

[0033] Acquisition module: used to collect the transient zero-sequence voltage signal of the target feeder as the power signal to be decomposed;

[0034] Decomposition module: used to perform variational mode decomposition on the electric quantity signal to be decomposed;

[0035] Detection module: Based on the one-dimensional feature vector of the electrical quantity signal to be decomposed obtained by the decomposition module, the fault detection result is calculated.

[0036] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned method for detecting high-resistance grounding faults in a distribution network are implemented.

[0037] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0038] Based on variational mode decomposition and Transformer network, the present invention proposes a new distribution network fault identification method with high recognition rate and robustness, and can accurately identify high-resistance grounding faults and disturbance conditions. The fault identification accuracy is very high, which improves the efficiency of distribution network fault detection and has important practical significance for preventing the occurrence of forest fire disasters.

[0039] Specifically, the present invention first builds a 10kV distribution network simulation model, and then collects a large amount of transient zero-sequence voltage waveform data of high-resistance grounding faults and normal disturbance conditions by changing system parameters and fault parameters. The collected signals are decomposed into different frequency modes using the VMD method, and finally the signal decomposition results are used as the input of the Transformer network to perform in-depth feature learning training on the transient signal, and then the trained model is used to effectively distinguish between HIF faults and disturbance conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0041] Figure 1 It is the overall structure diagram of the Transformer network model in the present invention;

[0042] Figure 2 A schematic diagram of the structure of a power distribution network used in an embodiment of the present invention;

[0043] Figure 3 is a diagram of an Emanuel arc model in an embodiment of the present invention;

[0044] Figure 4 It is a VMD algorithm decomposition result diagram in an embodiment of the present invention;

[0045] Figure 5 The accuracy rate and loss function of the model training set proposed in the embodiment of the present invention;

[0046] Figure 6 : is the test set confusion matrix of the three models in the embodiment of the present invention;

[0047] Figure 7 is the confusion matrix of the proposed model after adding noise in the embodiment of the present invention;

[0048] Figure 8 4 is a module diagram of a distribution network high-resistance grounding fault detection system in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with examples and drawings. The illustrative embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention. If specific conditions are not specified in the examples, they are carried out according to conventional conditions or conditions recommended by the manufacturer. If the manufacturer is not specified for the reagents or instruments used, they are all conventional products that can be purchased commercially.

[0050] A method for detecting a high-resistance grounding fault in a distribution network comprises the following steps:

[0051] S1, collecting the transient zero-sequence voltage signal of the target feeder as the electric quantity signal to be decomposed;

[0052] S2. performing variational modal decomposition on the electric quantity signal to be decomposed, wherein the modal functions obtained by decomposing a single electric quantity signal to be decomposed are arranged to obtain a one-dimensional feature vector of the electric quantity signal to be decomposed;

[0053] S3. Input the one-dimensional feature vector of the electric quantity signal to be decomposed into the pre-built Transformer model to obtain the detection result.

[0054] S1 can easily determine the fault type and fault degree by collecting zero-sequence voltage signals;

[0055] S2 can improve the signal noise reduction effect by performing VMD processing on the decomposed signal. VMD determines the frequency center and bandwidth of each component by iteratively searching for the optimal solution of the variational model, so as to adaptively realize the frequency domain decomposition of the signal and the effective separation of each component. This process helps to separate the noise and useful information in the signal and achieve the effect of noise reduction. The original signal can also be decomposed into multiple modal functions (IMFs) with limited bandwidth and center frequency. Each IMFs represents a relatively stable component in the signal to improve the accuracy and timeliness of fault detection. After the signal is decomposed into IMFs of different frequency components, they are arranged from low to high according to the frequency, so that the composition of different frequency components in the signal can be intuitively understood. This arrangement makes the low-frequency component in the front and the high-frequency component in the back, thereby improving the interpretability of signal analysis and helping to reveal the dynamic characteristics and change laws of the signal at different scales.

[0056] S3 analyzes and calculates the feature vector to obtain a fault detection structure with high accuracy and fast recognition efficiency.

[0057] Preferably, the modal functions obtained by decomposing the single electrical quantity signal to be decomposed in the above step S2 are arranged in order from low to high center frequencies.

[0058] Preferably, the zero-sequence voltage signal is a signal of one cycle before the zero-sequence voltage fault and two cycles after the fault at the head of each feeder. By comparing the changes in the zero-sequence voltage signal before and after the fault, the type of fault can be preliminarily determined; and the zero-sequence voltage reaches the highest value at the fault point, and gradually decreases as the distance from the fault point increases. Therefore, by collecting the zero-sequence voltage signal at the head of each feeder and combining the law of signal attenuation, the location range of the fault can be roughly determined. The zero-sequence voltage signal after the fault contains rich fault information, such as fault type, fault degree, etc.

[0059] In some embodiments of the present invention, the above step S2 includes:

[0060] The quantity signal to be decomposed is preset as f, and the quantity signal to be decomposed is decomposed into K modal components μ according to variational modal decomposition. k , the center frequency of the modal classification is ω k , taking the sum of the decomposed modal functions equal to the input signal as the constraint condition, the constrained variational equation is obtained, and then the constrained variational equation is transformed into an unconstrained variational equation according to the Lagrangian operator λ and the quadratic penalty factor α. Then, the alternating direction multiplier method is used to alternately iteratively update u(k) and ω(k) to find the saddle point of the unconstrained variational equation, thereby obtaining the optimal solution of the constrained variational model.

[0061] In some embodiments of the present invention, the above objective function and constraint conditions are respectively shown in formula (1) and formula (2):

[0062]

[0063]

[0064] Among them, {u k}={u1,u2,…,u K} is the kth modal component after decomposition, f(t) is the original signal, || ||2 represents the 2-norm of the orientation quantity, is the partial derivative of the function at time t, {ω k}={ω1,ω2,…,ω K} is the center frequency of each modal component, δ(t) is the unit impulse function, t represents the time variable, * is the convolution operator, and j represents the imaginary unit.

[0065] In some embodiments of the present invention, the constrained variational equation of equation (1) is converted into an unconstrained variational equation according to the Lagrangian operator λ and the quadratic penalty factor α, as shown in the following equation:

[0066]

[0067] Among them, λ(t) is the Lagrangian operator, α is the quadratic penalty factor, < > represents the inner product operation, represents the sum of all modal components.

[0068] In some embodiments of the present invention, the alternating direction multiplier method is used to iteratively update u(k) and ω(k) to find the saddle point of the unconstrained variational equation of formula (3), thereby obtaining the optimal solution of the constrained variational model of formulas (1) and (2):

[0069]

[0070] in, and They are f(t) and u k (t), λ(t), and The function obtained after Fourier transform, n represents the number of iterations, and ω represents the conversion of the signal from time domain to frequency domain.

[0071] In some embodiments of the present invention, the above also includes step S0, using the one-dimensional feature vector of the electrical quantity signal to be decomposed as the input data of the Transformer network, dividing the input data set into a training set and a test set, using the training set to train the Transformer model, and using the test set to test the network.

[0072] By using the data set to train the network model, S0 can take advantage of the powerful sequence modeling capabilities of the Transformer network and accurately capture subtle differences in the signal, thereby achieving accurate identification of different fault types and improving the accurate identification and classification of faults.

[0073] Preferably, the ratio of the training set to the test set is 19:1.

[0074] Preferably, the step S3 specifically utilizes a trained Transformer network to effectively distinguish between fault and disturbance conditions.

[0075] In some embodiments of the present invention, the overall structure of the Transformer network model in step S3 is referred to as Figure 1 , including input layer, position embedding layer, addition layer, self-attention layer, fully connected layer and output layer.

[0076] In the above, the signal after VMD decomposition is used as the input of the Transformer network, a Transformer network for HIF recognition is constructed, the network model is trained, the output result of the fully connected layer is extracted as the fault feature of the distribution line, and the Softmax classifier is used to effectively distinguish between fault and disturbance conditions.

[0077] In some embodiments of the present invention, in the above step S3

[0078] Input layer: The VMD algorithm is used to adaptively decompose the transient zero-sequence voltage signal, and the decomposition result is converted into one-dimensional time series data as the input of the Transformer network;

[0079] Position embedding layer: Since Transformer has no built-in sequence order information, the input features are mapped to a 256-dimensional latent space through the position embedding layer, and a learnable position encoding is introduced to enhance the representation of the position in the sequence;

[0080] Addition layer: The features of the input layer and the position embedding layer are integrated to enable the model to use both the content information and the position information of the input sequence, thereby enhancing the expressiveness and effect of the self-attention mechanism.

[0081] Self-attention layer: A total of two self-attention layers are used to calculate the attention weight corresponding to each position. The first layer uses causal masks to ensure that only the current position and the information before it can be seen at each position, which can effectively prevent information leakage. The second self-attention layer further enhances the model's global understanding of the entire sequence and its ability to learn complex features. Each layer is a self-attention layer with 4 heads and 128 keys;

[0082] Fully connected layer: The high-order features extracted by the self-attention layer are linearly mapped to the binary classification results for classification output;

[0083] Output layer: In this model, the disturbance conditions are classified into one category, so the model performs a binary classification task. The number of neurons is 2. The input of the output layer is the output of the fully connected layer, which is used as the output of this layer in the form of a probability vector.

[0084] In some embodiments of the present invention, the output layer adopts a Softmax classifier.

[0085] The Softmax classifier converts the output results of the fully connected layer into a probability distribution, so that the system can clearly determine whether the current state is a fault or a normal disturbance, facilitating efficient classification.

[0086] In some embodiments of the present invention, each of the above-mentioned self-attention layers is a self-attention layer with 4 heads and 128 keys.

[0087] In some embodiments of the present invention, the disturbance conditions are classified into one category in the model of the above-mentioned output layer, so the model performs a binary classification task, the number of neurons is 2, and the input of the output layer is the output of the fully connected layer, which is used as the output of this layer in the form of a probability vector.

[0088] In some embodiments of the present invention, the output result of the fully connected layer is extracted as the fault feature of the distribution line, and then the trained Transformer network is used to effectively distinguish between fault and disturbance conditions.

[0089] The present invention also provides a distribution network high-resistance grounding fault detection system, referring to Figure 8 ,include

[0090] Acquisition module 1: used to collect the transient zero-sequence voltage signal of the target feeder as the power signal to be decomposed;

[0091] Decomposition module 2: used for performing variational mode decomposition on the electric quantity signal to be decomposed;

[0092] Detection module 3: Based on the one-dimensional feature vector of the electric quantity signal to be decomposed obtained by decomposition module 2, a fault detection result is calculated.

[0093] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned method for detecting high-resistance grounding faults in a distribution network are implemented.

[0094] Example 1

[0095] In order to prove the effectiveness of the proposed method, a simulation platform was built on the PSCAD / EMTDC Figure 2The 10kV distribution network shown in the figure is used to simulate and analyze high-resistance grounding fault and normal disturbance conditions. 11 ~F 43 Indicates the fault location, L1~L4; the load size of each feeder is 0.6+j0.3MV·A.

[0096] The positive-sequence and zero-sequence impedance parameters of overhead lines and cables are shown in equations (6) and (7) respectively:

[0097]

[0098] The embodiment of the present invention uses the Emanuel arc model as the HIF fault model. The model is an empirical model constructed after a large number of field tests, has the advantages of simple parameter adjustment and high similarity with the actual fault model. Figure 3 shown.

[0099] The simulation step is 10 μs and the sampling step is 100 μs. The zero-sequence voltage waveforms of one cycle before and two cycles after the fault and disturbance conditions are intercepted as fault feature quantities. The simulation of HIF and disturbance conditions with different fault locations, different fault closing angles, and different fault resistance sizes is set. Only one of the parameters is changed each time for cyclic simulation, and the sample set shown in Table 1 is obtained.

[0100] Table 1 Simulation sample set

[0101]

[0102]

[0103] In order to verify the effectiveness of the VMD algorithm, a set of waveforms of HIF and disturbance conditions are randomly selected from the waveform database for testing. For K in VMD, it is 5, and the secondary penalty factor is 2500. The modal decomposition results of the zero-sequence voltage waveforms of high-resistance grounding fault and disturbance conditions are shown in Figure 4 shown.

[0104] Depend on Figure 4 It can be seen that various waveforms that are difficult to distinguish in the time domain are decomposed into modes with different center frequencies through VMD, which greatly manifests the characteristics of the original waveforms and improves the recognizability of the waveforms under different working conditions.

[0105] The sample set obtained in Table 1 is divided into training set and test set in a ratio of 19:1. The proposed model is tested after training and compared with the traditional support vector machine model and naive Bayes model. The training set accuracy and loss function curve of the proposed model are shown in Figure 5 The classification confusion matrix of the three models is shown in Figure 6 shown.

[0106] observe Figure 5 , Figure 6 It can be seen that the proposed model has no misjudgment and the recognition accuracy is 100%; the support vector machine model can effectively identify the fault condition, but cannot accurately exclude the disturbance condition, with an recognition accuracy of 95.9%; and the performance of the naive Bayes model is even worse, with an recognition accuracy of only 60%. This shows that the proposed model has completed the training and can accurately identify high-resistance grounding faults and various disturbance conditions based only on zero-sequence voltage.

[0107] The sample sets obtained in Table 1 are sufficient in number and free of noise interference, but this is not the case in reality. For actual distribution networks, fault samples are difficult to obtain in batches and are easily affected by noise during the communication process. Therefore, simulations are conducted on two situations: sample reduction and waveform interference with noise to verify the applicability of the proposed model.

[0108] In order to verify the performance of the proposed model under the condition of sample reduction, the samples obtained in Table 1 were reduced by 2 times, 4 times and 10 times respectively, and then the three models were trained and tested respectively. Among them, the proportion of the test set was 5%, 10% and 20% under different reduction factors. The obtained HIF recognition results are shown in Table 2.

[0109] Table 2 Fault recognition accuracy of three models after sample reduction

[0110]

[0111] From the results in Table 2, we can see that the proposed model is less affected by sample reduction, and even when the sample is reduced by 10 times, that is, 690 groups of samples, it still has a recognition accuracy of 97.10%; while the support vector machine model performs well when the samples are sufficient, but in the extreme case of 10 times reduction of samples, the fault recognition accuracy drops significantly.

[0112] In order to verify the influence of noise on the fault recognition accuracy of the proposed model, Gaussian white noise with signal-to-noise ratio of 20dB and 40dB was added to the test set of samples obtained in Table 1, and then the trained model was used for testing. The confusion matrix of the proposed model test set is as follows: Figure 7 shown.

[0113] Depend on Figure 7 It can be obtained that Gaussian white noise with a signal-to-noise ratio of 20dB and 40dB has almost no effect on the fault recognition accuracy of the proposed model, indicating that the proposed model can cope with the situation where the fault waveform in the actual distribution network is interfered by noise.

[0114] The detection method of the present invention has a high recognition rate and robustness, can accurately identify high-resistance grounding faults and disturbance conditions, has a very high fault recognition accuracy, improves the efficiency of distribution network fault detection, and has important practical significance for preventing the occurrence of wildfire disasters.

[0115] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for detecting high-resistance grounding faults in a distribution network, characterized in that: The steps include: S1, collecting the transient zero-sequence voltage signal of the target feeder as the electric quantity signal to be decomposed; S2. performing variational modal decomposition on the electric quantity signal to be decomposed, wherein the modal functions obtained by decomposing a single electric quantity signal to be decomposed are arranged to obtain a one-dimensional feature vector of the electric quantity signal to be decomposed; S3. Input the one-dimensional feature vector of the electric quantity signal to be decomposed into the pre-built Transformer model to obtain the detection result.

2. A method for detecting high-resistance grounding faults in a distribution network according to claim 1, characterized in that: The step S2 includes: presetting the electric quantity signal to be decomposed as f, decomposing the electric quantity signal to be decomposed into K modal components μ according to variational modal decomposition k , the center frequency of the modal classification is ω k , taking the sum of the decomposed modal functions equal to the input signal as the constraint condition, the constrained variational equation is obtained, and then the constrained variational equation is transformed into an unconstrained variational equation according to the Lagrangian operator λ and the quadratic penalty factor α. Then, the alternating direction multiplier method is used to alternately iteratively update u(k) and ω(k) to find the saddle point of the unconstrained variational equation, thereby obtaining the optimal solution of the constrained variational model.

3. A method for detecting high-resistance grounding faults in a distribution network according to claim 2, characterized in that: The objective function and constraint conditions are shown in formula (1) and formula (2) respectively: Among them, {u k }={u1,u2,…,u K } is the kth modal component after decomposition, f(t) is the original signal, 2 represents the 2-norm of the orientation quantity, is the partial derivative of the function at time t, {ω k }={ω1,ω2,…,ω K } is the center frequency of each modal component, δ(t) is the unit impulse function, t represents the time variable, * is the convolution operator, and j represents the imaginary unit.

4. A method for detecting high-resistance grounding faults in a distribution network according to claim 2, characterized in that: The unconstrained variational equation is shown as follows: Among them, λ(t) is the Lagrangian operator, α is the quadratic penalty factor, and represents the inner product operation. represents the sum of all modal components.

5. A method for detecting high-resistance grounding faults in a distribution network according to claim 2, characterized in that: Using the alternating direction multiplier method, u(k) and ω(k) are updated alternately and iteratively to find the saddle point of the unconstrained variational equation. The optimal solution of the constrained variational model is obtained by the following formula: in, and They are f(t) and u k (t), λ(t), and The function obtained after Fourier transform, n represents the number of iterations, and ω represents the conversion of the signal from time domain to frequency domain.

6. A method for detecting high-resistance grounding faults in a distribution network according to claim 1, characterized in that: The method also includes step S0, taking the one-dimensional feature vector of the electric quantity signal to be decomposed as the input data of the Transformer network, dividing the input data set into a training set and a test set, training the Transformer model with the training set, and testing the network with the test set.

7. A method for detecting high-resistance grounding faults in a distribution network according to claim 1, characterized in that: The overall structure of the Transformer network model in step S3 includes an input layer, a position embedding layer, an addition layer, a self-attention layer, a fully connected layer and an output layer.

8. A method for detecting high-resistance grounding faults in a distribution network according to claim 7, characterized in that: The output layer uses a variational mode decomposition algorithm to adaptively decompose the transient zero-sequence voltage signal, converts the decomposition result into one-dimensional time series data, and serves as the input of the Transformer network; The position embedding layer maps the input features to a 256-dimensional latent space through the position embedding layer, and introduces a learnable position encoding to enhance the representation of the position in the sequence; The addition layer fuses the features of the input layer and the position embedding layer, so that the model can simultaneously utilize the content information and position information of the input sequence, thereby enhancing the expressiveness and effect of the self-attention mechanism; The self-attention layer uses two layers of self-attention layers to calculate the attention weight corresponding to each position. The first layer uses a causal mask to ensure that only the current position and the information before it can be seen at each position. The second layer of self-attention layer further enhances the model's global understanding of the overall sequence and its ability to learn complex features. Each layer is a self-attention layer with 4 heads and 128 keys. The fully connected layer linearly maps the high-order features extracted by the self-attention layer to the binary classification results for classification output; The output layer classifies the disturbance conditions into one category, thus performing a binary classification task. The number of neurons is 2, and the input of the output layer is the output of the fully connected layer, which is output as a probability vector.

9. A distribution network high-resistance grounding fault detection system, characterized in that: include Acquisition module: used to collect the transient zero-sequence voltage signal of the target feeder as the power signal to be decomposed; Decomposition module: used to perform variational mode decomposition on the electric quantity signal to be decomposed; Detection module: Based on the one-dimensional feature vector of the electrical quantity signal to be decomposed obtained by the decomposition module, the fault detection result is calculated.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for detecting a high-resistance grounding fault in a distribution network as claimed in any one of claims 1 to 8 are implemented.

Images