Voltage sag disturbance source location and disturbance cause identification method and system

Abstract

The present application provides a method and system for locating a voltage sag disturbance source and identifying a disturbance cause. The method for locating a voltage sag disturbance source and identifying a disturbance cause comprises: collecting an input signal, wherein the input signal comprises five voltage data based on normal operation, single-phase grounding, two-phase grounding, three-phase grounding and two-phase short circuit; denoising the input signal based on a soft threshold wavelet transform, and obtaining a denoised signal; extracting features of the denoised signal by serially combining a CNN and an attention mechanism; inputting the extracted features into an attention mechanism module for adaptive weighting; inputting the feature extraction results into two fully connected layers for classification, and obtaining classification results, wherein the classification results comprise normal operation, single-phase grounding, two-phase grounding, three-phase grounding and two-phase short circuit, so as to solve the technical problem that the power system cannot identify and diagnose the voltage sag source in complex scenarios.

Description

Technical Field

[0001] The present application relates to the technical field of power quality analysis and detection methods, and in particular to a method and system for locating a voltage sag disturbance source and identifying the cause of the disturbance. Background Art

[0002] With the development of power electronics technology and information technology, the technical level of electrical equipment has been continuously improved. Many high-efficiency and high-performance equipment are widely used in many aspects of industrial production and people's daily life. However, most of these high-tech equipment are sensitive to changes in power supply characteristics. Among the power quality issues, voltage sag is one of the most prominent problems and has gradually become the focus of prevention and control.

[0003] Voltage sag refers to the power quality problem that the effective value of the power supply voltage drops rapidly to 90% to 10% of the rated value, lasts for 0.5 to 30 cycles, and then returns to the normal value. Serious power quality problems will slow down economic development and even cause great economic losses. At the same time, it is not conducive to the safe operation of the power grid. The harm of power quality is undoubtedly something that needs to be avoided by the country or individuals. Therefore, it is very important to locate the source of voltage sag disturbance and identify the cause of disturbance.

[0004] At present, the methods for identifying the source of voltage sag disturbances include methods based on mathematical models and methods based on signal processing. Among them, the method based on mathematical models needs to establish a mathematical model of the system, analyze the changes in voltage and current signals before and after the voltage sag occurs, and identify and diagnose the source of voltage sag. However, due to the complexity and uncertainty of the power system, it is very difficult to establish an accurate mathematical model. The method based on signal processing mainly uses digital signal processing technologies such as Fourier transform, Hilbert transform and wavelet transform to extract features of voltage sag disturbances, and then completes the identification of disturbance sources through classifiers. However, this method is more difficult to extract features when the scene is more complex, resulting in the performance of the classifier being easily affected by feature selection. Summary of the invention

[0005] The present application provides a method and system for locating a voltage sag disturbance source and identifying the disturbance cause, so as to solve the technical problem that a power system cannot identify and diagnose a voltage sag source in a complex scenario.

[0006] The present application provides a method for locating a voltage sag disturbance source and identifying a disturbance cause, comprising:

[0007] Collecting input signals, the input signals including five voltage data based on normal operation, single-phase grounding, double-phase grounding, three-phase grounding and double-phase short circuit;

[0008] De-noising the input signal based on soft threshold wavelet transform and obtaining a de-noised signal;

[0009] The CNN and attention mechanism are serially combined to extract features from denoised signals;

[0010] Input the extracted features into the attention mechanism module for adaptive weighting;

[0011] The feature extraction results are input into two fully connected layers for classification, and classification results are obtained, where the classification results include normal operation, single-phase grounding, double-phase grounding, three-phase grounding and double-phase short circuit.

[0012] Optionally, the step of denoising the input signal based on soft threshold wavelet transform and obtaining the denoised signal comprises the following steps:

[0013] The input signal is decomposed into 7 layers based on the 4th-order Daubechies wavelet as the basis function;

[0014]

[0015] Where x(n) represents the input signal, n = 0, 1, 2, ..., N-1, N represents the length of the input signal, C j (k) represents the low-frequency coefficient of the jth layer, D j (k) represents the high-frequency coefficient of the jth layer, h(n) and g(n) represent the low-pass and high-pass filter coefficients respectively, and k represents the index at each level.

[0016] Select a preset threshold, the preset threshold is:

[0017]

[0018] Where λ represents the threshold, μ1 and μ2 represent the parameters for adjusting the median and variance ratio, N is the input signal length, q is the wavelet threshold hyperparameter, median(·) represents the median operation, and D j (k) represents the high-frequency coefficient of the jth layer, k represents the index at each level,

[0019] Wavelet denoising is performed based on the soft threshold denoising formula. The soft threshold denoising formula is:

[0020]

[0021] in, is the updated wavelet coefficient, λ represents the threshold, D j (k) represents the high-frequency coefficient of the jth layer, k represents the index at each level,

[0022] The denoised signal is reconstructed by wavelet and the denoised signal is obtained.

[0023] Optionally, in the step of selecting a preset threshold, the preset threshold is updated to obtain an updated threshold:

[0024]

[0025] Among them, λ represents the threshold, is the updated threshold, and N is the signal length.

[0026] Optionally, the step of extracting features from the denoised signal by serially combining CNN and attention mechanism includes:

[0027] Use two convolutional layers for preliminary feature extraction;

[0028] The extracted features are input into the self-attention mechanism, and the local and global information of the input features are obtained.

[0029] Optionally, in the step of performing preliminary feature extraction using two convolutional layers, a preliminary voltage feature signal is obtained:

[0030] F i+1 =Conv 3×1 (Conv 3×1 (F i ))

[0031] Among them, F i is the input feature, F i+1 is the output feature, Conv represents the convolution operation, and i represents the output of the i-th layer network.

[0032] Optionally, the step of inputting the extracted features into the self-attention mechanism and obtaining local information and global information of the input features includes:

[0033] Based on the extracted features, the query Q, key K and value V in the attention mechanism are defined, and the local voltage feature information is obtained;

[0034] K=Conv 3×1 (F i )

[0035] V=Conv 1×1 (F i )

[0036] Among them, K represents the key, V represents the value, Conv represents the convolution operation, and F i is the input feature, and i represents the output of the i-th layer network.

[0037] Optionally, the step of inputting the extracted features into the attention mechanism module for adaptive weighting includes:

[0038] The query Q in the attention mechanism is concatenated with the key K, and the concatenated feature is input into two consecutive 1×1 convolutions to generate the attention matrix A.

[0039] A=Conv 1×1 (Conv 1×1 (Concat(Q,K)))

[0040] Among them, A represents the attention matrix, Concat represents the connection operation, Conv represents the convolution operation, Q represents the query, and K represents the key.

[0041] Multiply the attention matrix A with the numerical feature V and use the Softmax operation to obtain global information modeling;

[0042] Weight the output results and the key K matrix to combine local information with global information;

[0043]

[0044] in, represents the element addition operation, Represents the element-wise multiplication operation, F out Represents the output of the attention mechanism, A represents the attention matrix, V represents the value, and K represents the key.

[0045] Optionally, the feature extraction result is input into two fully connected layers for classification, and the classification result is obtained, wherein the classification result includes the steps of normal operation, single-phase grounding, two-phase grounding, three-phase grounding and two-phase short circuit.

[0046] The obtained feature results are input into two fully connected layers to complete the classification, and ReLU is selected as the activation function. The output is:

[0047] F out =ReLU(W i F i +b i )

[0048] Among them, W i is the weight matrix, b i is the bias vector, F i is the input feature, F out is the output result.

[0049] Correspondingly, the present application also provides an electronic device, which includes a memory and a processor, the memory being used to store executable program code; the processor being connected to the memory, and running a computer program corresponding to the executable program code by reading the executable program code, so as to execute the steps in any of the aforementioned methods for locating the source of voltage sag disturbance and identifying the cause of disturbance.

[0050] Correspondingly, the present application also provides a method system for locating a voltage sag disturbance source and identifying a disturbance cause, characterized in that it includes the electronic device described above.

[0051] The present application provides a method and system for locating the source of voltage sag disturbance and identifying the cause of disturbance, which uses wavelet packet decomposition on the input signal, uses wavelet transform to decompose the signal into detail information of different frequencies, and then performs threshold denoising on the decomposed wavelet coefficients. When selecting the threshold, a soft threshold processing method is used, and a combination of variance and median is used in the processing to perform threshold adaptive determination, and finally reconstruct the signal. This method can efficiently process noise data and improve recognition accuracy.

[0052] The convolutional neural network model that combines CNN and self-attention mechanism combines the advantages of convolutional neural network and attention mechanism. It extracts input features by using CNN convolution, and then inputs the extracted features into the attention mechanism to adaptively adjust the importance of each channel. It further screens the features to adapt to different voltage sag disturbances and improve the robustness of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0054] Figure 1 It is a flow chart of the method for locating the voltage sag disturbance source and identifying the disturbance cause provided by the present application;

[0055] Figure 2 It is a flowchart of step S200 in the method for locating the voltage sag disturbance source and identifying the disturbance cause provided by the present application;

[0056] Figure 3 It is a schematic diagram of the simulation structure of the voltage sag disturbance source location and disturbance cause identification system provided in this application. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application. In addition, it should be understood that the specific implementation methods described herein are only used to illustrate and explain the present application and are not used to limit the present application. In the present application, unless otherwise stated, the directional words used, such as "up", "down", "left", and "right", generally refer to the up, down, left, and right of the device in actual use or working state, specifically the drawing direction in the accompanying drawings.

[0058] The present application provides a method and system for locating a voltage sag disturbance source and identifying the cause of the disturbance, which are described in detail below. It should be noted that the description order of the following embodiments is not intended to limit the preferred order of the embodiments of the present application. In the following embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0059] See also Figure 1 The present application provides a method for locating the source of voltage sag disturbance and identifying the cause of disturbance. The method first denoises the input signal through wavelet transform, and inputs the denoised signal into a convolutional neural network to complete the location of the voltage sag disturbance source and the identification of the disturbance cause.

[0060] See also Figure 1 and Figure 2 A method for locating a voltage sag disturbance source and identifying a disturbance cause comprises the following steps:

[0061] S100, collecting input signals, wherein the input signals include five types of voltage data based on normal operation, single-phase grounding, two-phase grounding, three-phase grounding and two-phase short circuit;

[0062] Five types of voltage data are collected based on normal operation, single-phase grounding, two-phase grounding, three-phase grounding and two-phase short circuit, and the data are randomly divided into training set and test set according to the sample ratio of 7:3.

[0063] S200, performing denoising on the input signal based on soft threshold wavelet transform, and obtaining a denoised signal;

[0064] In real life, the voltage signal obtained may contain noise. Wavelet transform denoising, as a common signal processing method, can effectively filter out the noise in the signal, thereby improving the quality and reliability of the signal. The specific operations are as follows:

[0065] S210, decomposing the voltage disturbance signal into 7 layers based on selecting the 4th-order Daubechies wavelet as the basis function, and for the decomposed signal, the wavelet coefficient of the noise is smaller than the wavelet coefficient of the signal;

[0066] For an input signal x(n) of length N, where n = 0, 1, 2, ..., N-1, the discrete wavelet transform can be implemented by the following steps,

[0067]

[0068] Where x(n) represents the input signal, n = 0, 1, 2, ..., N-1, N represents the length of the input signal, C j (k) represents the low-frequency coefficient of the jth layer, D j (k) represents the high-frequency coefficient of the jth layer, h(n) and g(n) represent the low-pass and high-pass filter coefficients respectively, and k represents the index at each level.

[0069] S220, selecting a preset threshold;

[0070] By selecting a suitable threshold, the noise is removed. In the traditional soft threshold processing method, the threshold is selected by the following formula, where median(·) represents the median operation, and the threshold is selected by selecting the median of the main components in the signal.

[0071]

[0072] Among them, λ represents the threshold, median(·) represents the median operation, and D j (k) represents the high-frequency coefficient of the jth layer, k represents the index at each level,

[0073] In this embodiment, the variance is added to the original wavelet soft threshold calculation, and the variance can automatically adapt to the noise intensity in the signal. When the variance is large, the threshold will be increased accordingly to better suppress the noise. In order to ensure the adaptability and consistency of the threshold, the following formula is used to obtain the preset threshold:

[0074]

[0075] Where λ represents the threshold, μ1 and μ2 represent the parameters for adjusting the median and variance ratio, N is the input signal length, q is the wavelet threshold hyperparameter, median(·) represents the median operation, and D j (k) represents the high-frequency coefficient of the jth layer, k represents the index at each level,

[0076] Update the preset threshold and obtain the updated threshold:

[0077]

[0078] Among them, λ represents the threshold, is the updated threshold, and N is the signal length.

[0079] S230, performing wavelet denoising based on a soft threshold denoising formula, the soft threshold denoising formula is:

[0080]

[0081] in, is the updated wavelet coefficient, λ represents the threshold, D j (k) represents the high-frequency coefficient of the jth layer, k represents the index at each level,

[0082] S240, performing wavelet reconstruction on the denoised signal to obtain a denoised signal;

[0083] In the step of using wavelet transform to denoise the input signal, the input signal is decomposed using DB4 wavelet packet, and the signal is decomposed into detail information of different frequencies using wavelet transform, and then the decomposed wavelet coefficients are threshold denoised. When selecting the threshold, soft threshold processing is used, and the variance and median are combined to adaptively determine the threshold, and finally the signal is reconstructed. This method can efficiently process noisy data and improve recognition accuracy.

[0084] S300, uses CNN and attention mechanism to extract features from denoised signals in series;

[0085] The convolutional neural network based on the self-attention mechanism is divided into two parts. The first part is the preliminary feature extraction. In this part, two convolutional layers are used for preliminary feature extraction, and ReLU is selected as the activation function. The second part uses the self-attention mechanism to combine the local and global information of the input features to further improve the classification accuracy and robustness of the model.

[0086] In the step of using two convolutional layers for preliminary feature extraction, a preliminary voltage feature signal is obtained:

[0087] F i+1 =Conv 3×1 (Conv 3×1 (F i ))

[0088] Among them, F i is the input feature, F i+1 is the output feature, Conv represents the convolution operation, and i represents the output of the i-th layer network.

[0089] Since the receptive field of the convolutional neural network constructed by CNN is small, it is difficult to further extract the global information of the features. Therefore, in the second part of this algorithm, the self-attention mechanism is used to combine the local and global information of the input features to further improve the classification accuracy and robustness of the model. The global information can help the model focus on the overall information and capture the overall distribution characteristics of the world, while the local information can focus on key local areas, such as voltage mutation information, and capture local detail features. The attention mechanism algorithm is introduced as follows:

[0090] Based on the extracted features, the query Q, key K and value V in the attention mechanism are defined, and the local voltage feature information is obtained;

[0091] K=Conv 3×1 (F i )

[0092] V=Conv 1×1 (F i )

[0093] Among them, K represents the key, V represents the value, Conv represents the convolution operation, and F i is the input feature, i represents the output of the i-th layer network;

[0094] First, for the input features, three variables Q, K, and V are defined to represent the query Q, key K, and value V in the attention mechanism, respectively. V represents the numerical information in the input sequence, and the attention weights obtained by Q and K are then multiplied by the value V to implement the attention mechanism. K and V use 3×1 and 1×1 convolutions to complete the feature mapping, respectively. In the above formula, K obtains local voltage feature information because it uses 3×1 convolution. First, for the input features, three variables Q, K, and V are defined to represent the query Q, key K, and value V in the attention mechanism, respectively. V represents the numerical information in the input sequence, and the attention weights obtained by Q and K are then multiplied by the value V to implement the attention mechanism. K and V use 3×1 and 1×1 convolutions to complete the feature mapping, respectively. In the above formula, K obtains local voltage feature information because it uses 3×1 convolution.

[0095] S400, input the extracted features into the attention mechanism module for adaptive weighting;

[0096] Step S400 specifically includes the following steps:

[0097] The query Q in the attention mechanism is concatenated with the key K, and the concatenated feature is input into two consecutive 1×1 convolutions to generate the attention matrix A.

[0098] A=Conv 1×1 (Conv 1×1(Concat(Q,K)))

[0099] Among them, A represents the attention matrix, Concat represents the connection operation, Conv represents the convolution operation, Q represents the query, and K represents the key;

[0100] Different from the traditional self-attention mechanism, the attention matrix generated at this time is obtained by the interaction of input feature information and local information, rather than just modeling the relationship between Q and K.

[0101] Multiply the attention matrix A with the numerical feature V and use the Softmax operation to obtain global information modeling;

[0102] Weight the output results and the key K matrix to combine local information with global information;

[0103]

[0104] in, represents the element addition operation, Represents the element-wise multiplication operation, F out represents the output of the attention mechanism, A represents the attention matrix, V represents the value, and K represents the key;

[0105] The convolutional neural network model that combines CNN and self-attention mechanism combines the advantages of convolutional neural network and attention mechanism. It extracts input features by using CNN convolution, and then inputs the extracted features into the attention mechanism to adaptively adjust the importance of each channel. It further screens the features to adapt to different voltage sag disturbances and improve the robustness of the model.

[0106] S500, inputting the feature extraction results into two fully connected layers for classification, and obtaining classification results, wherein the classification results include normal operation, single-phase grounding, two-phase grounding, three-phase grounding and two-phase short circuit;

[0107] In step S500, the obtained feature results are input into two fully connected layers to finally complete the classification, and ReLU is selected as the activation function, and the output is:

[0108] F out =ReLU(W i F i +b i )

[0109] Among them, W i is the weight matrix, b i is the bias vector, F i is the input feature, F out is the output result.

[0110] In order to solve the challenges of tedious manual feature extraction and processing complex scenes, and to improve these problems without increasing the computational complexity, this embodiment proposes a method for locating the source of voltage sag disturbance and identifying the cause of disturbance by combining wavelet transform based on soft threshold and convolutional neural network based on self-attention mechanism. This method combines wavelet transform and convolutional neural network, has good stability, strong robustness to noise, and certain generalization ability.

[0111] The method for locating the source of voltage sag disturbance and identifying the cause of disturbance combines the advantages of wavelet transform denoising and convolutional neural network. It avoids the tedious manual feature extraction process while having relatively low computational complexity. It can locate the source of voltage sag disturbance and identify the cause of disturbance under certain noise interference.

[0112] In order to verify the rationality of the above method, we collected five different voltage information of three different nodes, including normal operation, single-phase grounding, two-phase grounding, three-phase grounding and two-phase short circuit information. The system structure diagram is shown in the figure. Figure 3 As shown, the short-circuited nodes are F1, F2 and F3 respectively. During the training process, batch_size is set to 256, epoch is 160, wavelet threshold hyperparameter is set to 0.95, median and variance coefficient are both set to 0.5, and the loss function uses the cross entropy loss function, as shown in the following formula:

[0113]

[0114] Among them, L represents the cross entropy loss function, y i represents the i-th element of the true label, p i represents the predicted probability of the model for the i-th category, and n represents the amount of voltage information.

[0115] The input data is normalized and input into the model, and 20db, 30db, and 40db Gaussian noise are added to the input. The experimental results are shown in Table 1.

[0116] Table 1 Experimental test results

[0117] original signal 20db noise 30db noise 40db noise Single fault accuracy 100％ 99.82％ 99.92％ 100％ Double fault accuracy 99.73％ 99.62％ 99.75％ 99.87％ Multiple fault accuracy 99.99％ 99.56％ 99.68％ 99.88％

[0118] A single fault means that only one of the three nodes will have a short circuit fault, a double fault means that two nodes will have a short circuit fault at the same time, and a multiple fault is composed of a single fault, a double fault, and a fault at all three nodes. The test data is sent to the trained network to locate the source of the voltage sag disturbance and identify the cause of the disturbance.

[0119] The present application also includes a voltage sag disturbance source location and disturbance cause identification system, which includes an electronic device. The electronic device includes a memory and a processor, the memory is used to store executable program code; the processor is connected to the memory, and runs a computer program corresponding to the executable program code by reading the executable program code to execute the steps in the above-mentioned voltage sag disturbance source location and disturbance cause identification method.

[0120] The above is a detailed introduction to a method and system for locating the source of voltage sag disturbance and identifying the cause of disturbance provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for a person skilled in the art, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for locating a voltage sag disturbance source and identifying the disturbance cause, characterized in that: include: Collecting input signals, the input signals including five voltage data based on normal operation, single-phase grounding, double-phase grounding, three-phase grounding and double-phase short circuit; De-noising the input signal based on soft threshold wavelet transform and obtaining a de-noised signal; The CNN and attention mechanism are serially combined to extract features from denoised signals; Input the extracted features into the attention mechanism module for adaptive weighting; Input the feature extraction results into two fully connected layers for classification, and obtain classification results, wherein the classification results include normal operation, single-phase grounding, double-phase grounding, three-phase grounding and double-phase short circuit; The step of performing denoising on the input signal based on soft threshold wavelet transform and obtaining the denoised signal comprises the following steps: The input signal is decomposed into 7 layers based on the 4th-order Daubechies wavelet as the basis function; Where x(n) represents the input signal, n = 0, 1, 2, ..., N-1, N represents the length of the input signal, C j (k) represents the low-frequency coefficient of the jth layer, D j (k) represents the high-frequency coefficient of the jth layer, h(n) and g(n) represent the low-pass and high-pass filter coefficients respectively, and k represents the index at each level. Select a preset threshold, the preset threshold is: Where λ represents the threshold, μ1 and μ2 represent the parameters for adjusting the median and variance ratio, N is the input signal length, q is the wavelet threshold hyperparameter, median(·) represents the median operation, and D j (k) represents the high-frequency coefficient of the jth layer, k represents the index at each level, Wavelet denoising is performed based on the soft threshold denoising formula. The soft threshold denoising formula is: in, is the updated wavelet coefficient, λ represents the threshold, D j (k) represents the high-frequency coefficient of the jth layer, k represents the index at each level, Perform wavelet reconstruction on the denoised signal and obtain the denoised signal; In the step of selecting a preset threshold, the preset threshold is updated and an updated threshold is obtained: Among them, λ represents the threshold, is the updated threshold, N is the signal length; Use two convolutional layers for preliminary feature extraction; In the step of using two convolutional layers to perform preliminary feature extraction, a preliminary voltage feature signal is obtained: F i+1 =Conv 3×1 (Conv 3×1 (F i )) Among them, F i is the input feature, F i+1 is the output feature, Conv represents the convolution operation, and i represents the output of the i-th layer network; The step of extracting features from the denoised signal by serially combining CNN and attention mechanism includes: Input the extracted features into the self-attention mechanism and obtain the local and global information of the input features; The step of inputting the extracted features into the self-attention mechanism and obtaining local information and global information of the input features includes: Based on the extracted features, the query Q, key K and value V in the attention mechanism are defined, and the local voltage feature information is obtained; K=Conv 3×1 (F i ) V=Conv 1×1 (F i ) Among them, K represents the key, V represents the value, Conv represents the convolution operation, and F i is the input feature, i represents the output of the i-th layer network; The step of inputting the extracted features into the attention mechanism module for adaptive weighting includes: The query Q in the attention mechanism is concatenated with the key K, and the concatenated feature is input into two consecutive 1×1 convolutions to generate the attention matrix A. A=Conv 1×1 (Conv 1×1 (Concat(Q,K))) Among them, A represents the attention matrix, Concat represents the connection operation, Conv represents the convolution operation, Q represents the query, and K represents the key. Multiply the attention matrix A with the numerical feature V and use the Softmax operation to obtain global information modeling; Weight the output results and the key K matrix to combine local information with global information; in, represents the element addition operation, Represents the element-wise multiplication operation, F out Represents the output of the attention mechanism, A represents the attention matrix, V represents the value, and K represents the key.

2. The method for locating the voltage sag disturbance source and identifying the disturbance cause according to claim 1, characterized in that: The feature extraction results are input into two fully connected layers for classification, and the classification results are obtained, wherein the classification results include normal operation, single-phase grounding, two-phase grounding, three-phase grounding and two-phase short circuit. The obtained feature results are input into two fully connected layers to complete the classification, and ReLU is selected as the activation function. The output is: F out =ReLU(W i F i +b i ) Among them, W i is the weight matrix, b i is the bias vector, F i is the input feature, F out is the output result.

3. An electronic device, characterized in that: include: A memory for storing executable program codes; as well as A processor is connected to the memory, and runs a computer program corresponding to the executable program code by reading the executable program code to execute the steps in the method for locating a voltage sag disturbance source and identifying a disturbance cause as claimed in any one of claims 1 to 2.

4. A system for locating voltage sag disturbance sources and identifying disturbance causes, characterized in that: Comprising the electronic device as claimed in claim 3.

Images