Power distribution network fault identification method, system, equipment and medium

Abstract

The invention relates to the technical field of power distribution network fault identification, and discloses a power distribution network fault identification method, system and device and a medium, and the method comprises the steps: obtaining and preprocessing a fault recording signal of a target power distribution network, and obtaining a second fault recording signal waveform; performing decomposition operation to obtain a first time-frequency matrix, and converting the first time-frequency matrix into a second time-frequency feature map; and performing sample expansion on the second time-frequency feature map to form a first fault sample set which comprises expanded and original feature maps. And establishing a fault identification model, and training the model by using the sample set. And after the training is completed, performing fault identification by using the model. According to the method, signal processing and quantization coding technologies are combined, the power distribution network fault is analyzed through the two-dimensional image, and the credibility of the fault signal is enhanced. Meanwhile, the fault sample data of the power distribution network is expanded by utilizing the generative adversarial network, so that the over-fitting of the model due to insufficient samples is prevented, and the recognition capability of the system on different faults is improved.

Description

Technical Field

[0001] The present application relates to the technical field of distribution network fault identification, and in particular to a distribution network fault identification method, system, device and medium. Background Art

[0002] The distribution network is a vital component of the power grid, transmitting electrical energy to consumers. When a distribution network fault occurs, the selection of the faulty line and protective action both rely on the accurate identification of the fault type. As the proportion of renewable energy connected to the grid increases, the disturbance of fault signals increases, placing higher demands on the accurate identification of distribution network faults.

[0003] There are two main methods for fault identification in distribution networks. The first is based on the analysis of steady-state characteristics of fault signals, and the other is based on the analysis of transient characteristics of fault signals. Fault identification is usually divided into two steps. The first step is to extract fault features, extracting electrical quantities or corresponding statistical information from distribution network fault signals as feature quantities for fault identification. The second step is to select the optimal machine learning algorithm to identify fault features, thereby realizing distribution network fault identification. Rapid and accurate acquisition of fault features is a prerequisite for the accuracy of fault type identification in distribution networks. The premise of extracting fault features is relatively complex and requires strong prior knowledge and manual experience. The access of renewable energy sources such as photovoltaic and wind power makes the fault signal exhibit non-stationary characteristics. Analysis based on electrical quantities cannot detect sudden changes in signal quantities. Deep learning can be used to solve this problem, thereby realizing adaptive extraction of fault features.

[0004] The existing technology proposes to use binary discrete wavelet transform to process signals, select deep neural network for training, and use sample error as the objective function for optimization. This method works well for ground faults, but the number of samples of different fault types is seriously unbalanced, which leads to overfitting when processing other fault types and makes it impossible to distinguish all fault types. Summary of the Invention

[0005] In view of the above existing problems, this application is proposed.

[0006] Therefore, the present application provides a distribution network fault identification method, system, device and medium, which can solve the overfitting problem caused by the imbalance in the number of samples of different fault types.

[0007] To solve the above technical problems, this application provides the following technical solutions:

[0008] In a first aspect, the present application provides a distribution network fault identification method, comprising:

[0009] Acquire a first fault recording signal waveform of a target distribution network, and perform a first preprocessing on the fault recording signal waveform to obtain a second fault recording signal waveform;

[0010] Performing a first decomposition operation on the second fault recording signal waveform to obtain a first time-frequency matrix;

[0011] Converting the first time-frequency matrix into a second time-frequency feature map;

[0012] Performing a sample expansion operation on the second time-frequency feature graph, and using the second time-frequency feature graph after sample expansion and the original second time-frequency feature graph as a first fault sample set;

[0013] Establishing a first fault identification model, and using the first fault sample set as a sample set of the first fault identification model for model training;

[0014] The target distribution network fault is identified based on the first fault identification model after training.

[0015] As a preferred solution of the distribution network fault identification method described in the present application, wherein: the first decomposition operation on the second fault recording signal waveform includes:

[0016] performing a first frequency band division on the waveform of the second fault recording signal to obtain a plurality of frequency bands;

[0017] Establishing a plurality of filters for each frequency band, and obtaining a plurality of fault filtering signals according to the plurality of filters;

[0018] A target fault filter signal waveform modal function is obtained according to the plurality of fault filter signals.

[0019] As a preferred solution of the distribution network fault identification method described in the present application, wherein: the performing a first decomposition operation on the second fault recording signal waveform further includes:

[0020] Performing a first sorting of the target fault filter signal waveform modal functions according to center frequencies;

[0021] and performing a second sorting on all target fault filter signal waveform modal functions;

[0022] A first time-frequency matrix is ​​obtained according to the first sorting and the second sorting results.

[0023] As a preferred solution of the distribution network fault identification method described in the present application, wherein: converting the first time-frequency matrix into a second time-frequency feature graph includes:

[0024] Performing quantization encoding on the first time-frequency matrix;

[0025] The quantization coding is used to convert the first time-frequency matrix into a pixel matrix;

[0026] The quantization coding converts the first time-frequency matrix into a pixel matrix within a target pixel interval range through a random rounding operation to obtain a second time-frequency feature map.

[0027] As a preferred solution of the distribution network fault identification method described in the present application, wherein: the sample expansion operation on the second time-frequency feature graph includes:

[0028] Determining a fault type label for generating a first fault recording signal waveform of a target distribution network, and performing a separate encoding operation on the fault type label;

[0029] The first sample expansion model is trained using the fault type label after the separate encoding operation and the second time-frequency feature map representing the current system working condition as the conditional value and the real sample respectively;

[0030] The output of the first sample expansion model is the second time-frequency feature graph after sample expansion, and the acquired second time-frequency feature graph during distribution network failure is the original second time-frequency feature graph.

[0031] This optimization scheme significantly improves the model's generalization capabilities, enabling the distribution network fault identification system to maintain high accuracy even when faced with new, unseen fault types. By encoding fault type labels individually, the model can learn the subtle differences between different fault types, allowing it to more accurately simulate a variety of possible fault scenarios during sample expansion.

[0032] As a preferred solution of the distribution network fault identification method described in this application, wherein: the establishing of the first fault identification model includes:

[0033] Compressing the first fault sample set to obtain a feature image of a target size;

[0034] Taking the characteristic image of the target size as input, and inputting it into an initially untrained first fault recognition model;

[0035] The first fault identification model includes an input layer, two sets of alternating convolutional layers and pooling layers, a fully connected layer, and an output layer.

[0036] As a preferred solution of the distribution network fault identification method described in the present application, wherein: the target distribution network fault identification based on the trained first fault identification model includes:

[0037] The first fault identification model further includes a classifier connected to the output layer;

[0038] The classifier is used to perform fault identification and classification on the output of the first fault identification model;

[0039] The process of establishing the classifier includes:

[0040] Calculating the mode of the first fault sample set, and calculating the similarity between the mode and each first fault sample;

[0041] determining a weight of each first fault sample according to the similarity;

[0042] Construct a classifier based on the first fault sample set after weight determination.

[0043] In a second aspect, the present application provides a distribution network fault identification system, comprising:

[0044] A data acquisition and processing module is used to acquire a first fault recording signal waveform of a target distribution network, and perform a first preprocessing on the fault recording signal waveform to obtain a second fault recording signal waveform;

[0045] a matrix acquisition module, configured to perform a first decomposition operation on the second fault recording signal waveform to obtain a first time-frequency matrix;

[0046] A feature map acquisition module, configured to convert the first time-frequency matrix into a second time-frequency feature map;

[0047] an expansion module, configured to perform a sample expansion operation on the second time-frequency characteristic graph, and use the second time-frequency characteristic graph after sample expansion and the original second time-frequency characteristic graph as a first fault sample set;

[0048] A model building module, configured to build a first fault identification model and use the first fault sample set as a sample set of the first fault identification model for model training;

[0049] The identification module is used to identify the target distribution network fault according to the first fault identification model after training.

[0050] In a third aspect, the present application provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described above when executing the computer program.

[0051] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method described above when the computer program is executed by a processor.

[0052] Compared with the prior art, the beneficial effects of the present application are as follows: the present application proposes a distribution network fault identification method, system, device and medium, obtains a first fault recording signal waveform of the target distribution network, and performs a first preprocessing on the fault recording signal waveform to obtain a second fault recording signal waveform; performs a first decomposition operation on the second fault recording signal waveform to obtain a first time-frequency matrix; converts the first time-frequency matrix into a second time-frequency feature graph; performs a sample expansion operation on the second time-frequency feature graph, and uses the second time-frequency feature graph after sample expansion and the original second time-frequency feature graph as a first fault sample set; establishes a first fault identification model, and uses the first fault sample set as a sample set of the first fault identification model for model training; and performs target distribution network fault identification based on the trained first fault identification model. This application combines signal processing technology with quantization coding to analyze distribution network faults from the perspective of two-dimensional images, thereby enhancing the unreliable information in the fault signal. At the same time, it proposes the use of a generative adversarial network to expand the distribution network fault sample data from the perspective of image expansion, thereby avoiding the overfitting problem of the fault identification model due to insufficient training sample data, and improving the system's ability to identify different faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0054] Figure 1 A flow chart of a method for identifying distribution network faults provided in one embodiment of the present application.

[0055] Figure 2 This is a diagram of the internal structure of an electronic device for a distribution network fault identification method provided in one embodiment of the present application. DETAILED DESCRIPTION

[0056] To make the above-mentioned purposes, features, and advantages of this application more clearly understood, the following detailed description of the specific embodiments of this application is given in conjunction with the accompanying drawings. It is obvious that the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of this application.

[0057] Example 1, with reference to Figure 1-Figure 2 , which is the first embodiment of the present application, provides a distribution network fault identification method, comprising:

[0058] There are some problems in the existing related technologies. For example, the number of samples of different fault types is unbalanced, which leads to overfitting when processing other fault types, limiting the generalization ability of the fault identification model.

[0059] This application provides a method that can effectively solve the above-mentioned problems. Next, we will describe in detail how to implement the distribution network fault identification method in combination with multiple embodiments.

[0060] Figure 1 A method flow chart of a distribution network fault identification method is shown, including:

[0061] S101, obtaining a first fault recording signal waveform of a target distribution network, and performing a first preprocessing on the fault recording signal waveform to obtain a second fault recording signal waveform;

[0062] It should be noted that the fault recording signal waveform is a graphical representation of the changes in electrical quantities recorded by a fault recording device (such as an FTU) installed in the target distribution network when a power system fault occurs. These electrical quantities include, but are not limited to, three-phase voltage, three-phase current, zero-sequence voltage, and zero-sequence current. A fault recorder can capture and record changes in electrical parameters over a period of time before and after a fault occurs, which is crucial for analyzing the cause, type, and location of the fault.

[0063] In an optional embodiment, the first preprocessing may include denoising, filtering, and standardizing the waveform of the first fault recording signal to remove noise interference, improve signal quality, and convert the signal into a unified standard format to facilitate subsequent processing and analysis. The denoising operation can use digital filtering techniques, such as low-pass filtering or band-pass filtering, to remove high-frequency noise components; the filtering operation selects an appropriate filter based on the characteristics of the distribution network fault signal; and the standardization process normalizes the signal parameters such as amplitude and frequency to make different fault signals comparable.

[0064] It should be noted that to ensure that all relevant waveforms are captured when a fault occurs in the target distribution network and to avoid wasting resources due to excessive redundant waveforms, a fault cycle can be set for waveform acquisition. For example, acquisition can begin some time before the fault occurs and continue some time after the fault occurs. This ensures that the fault waveform is fully recorded while avoiding unnecessary waste of resources. Furthermore, to ensure the accuracy of the waveform data, the quality of the collected first fault recording signal waveform can be assessed, and those signal waveforms with substandard quality, such as those with a low signal-to-noise ratio, can be eliminated to ensure the accuracy and reliability of subsequent processing.

[0065] In the embodiment of the present application, data of six cycles before a fault and eight cycles after a fault are collected each time as the fault signal to be decomposed and subjected to the first preprocessing.

[0066] In an embodiment of the present application, the first preprocessing removes abnormal data from the original data, uses interpolation to fill in the data, collects data according to 64 sampling points in each cycle, and parses data such as three-phase voltage, three-phase current, zero-sequence voltage and zero-sequence current from the preprocessed fault recording.

[0067] It should be noted that obtaining the first fault recording signal waveform of the target distribution network and performing a first preprocessing on the fault recording signal waveform to obtain the second fault recording signal waveform can improve the accuracy and reliability of fault identification. The preprocessed second fault recording signal waveform is clearer and has reduced noise interference, which is conducive to the subsequent extraction and analysis of fault features. At the same time, through standardization, different fault signals are made comparable, providing more accurate and reliable input data for subsequent machine learning algorithms, thereby improving the accuracy and efficiency of fault identification. In addition, preprocessing the fault recording signal waveform can also reduce the amount of calculation, improve the real-time performance of fault identification, and provide more timely and effective protection for the safe and stable operation of the power system.

[0068] S102, performing a first decomposition operation on the second fault recording signal waveform to obtain a first time-frequency matrix;

[0069] In an optional embodiment, the first decomposition operation is to decompose the complex fault recording signal waveform into a series of components with clear physical meanings, which can better reflect the fault characteristics.

[0070] In an optional embodiment, the first decomposition operation can use empirical mode decomposition (EMD) technology, which can decompose nonlinear and non-stationary signals into a series of intrinsic mode functions (IMFs), and each IMF component represents an inherent vibration mode of the signal. In this way, the characteristic information in the fault signal can be better extracted, providing a basis for subsequent time-frequency analysis. EMD technology has the characteristics of adaptability and locality. It can perform adaptive decomposition according to the characteristics of different fault signals to obtain IMF components with clear physical meanings, which is crucial for the accuracy of fault identification. At the same time, EMD technology can also retain the nonlinear characteristics of the signal, which is of great significance for processing complex distribution network fault signals.

[0071] In an optional embodiment, the first decomposition operation may also utilize wavelet transform technology, which can decompose the signal into wavelet components of different frequencies, thereby revealing the characteristics of the signal at different frequencies. Wavelet transforms have the capability of multi-resolution analysis, and different wavelet basis functions and decomposition levels can be selected as needed to achieve a detailed analysis of the fault signal. Wavelet transforms can obtain the time-frequency distribution characteristics of the signal, which is of great significance for capturing the instantaneous changes of the fault signal and can further improve the accuracy and sensitivity of fault identification.

[0072] It should be noted that single wavelet decomposition or single empirical mode decomposition cannot well meet the processing requirements of complex distribution network fault signals. Therefore, in the embodiment of the present application, a hybrid decomposition method combining empirical mode decomposition (EMD) and wavelet transform technology is preferably used.

[0073] In the embodiment of the present application, performing a first decomposition operation on the second fault recording signal waveform includes:

[0074] Dividing the second fault recording signal waveform into a first frequency band to obtain a plurality of frequency bands;

[0075] Establishing a plurality of filters for each frequency band, and obtaining a plurality of fault filtering signals according to the plurality of filters;

[0076] According to a plurality of fault filtering signals, a target fault filtering signal waveform modal function is obtained.

[0077] In the embodiment of the present application, performing the first decomposition operation on the second fault recording signal waveform further includes:

[0078] Sort the target fault filter signal waveform modal functions by center frequency;

[0079] and performing a second sorting on all target fault filter signal waveform modal functions;

[0080] A first time-frequency matrix is ​​obtained according to the first sorting and the second sorting results.

[0081] In an optional embodiment, the first and second sorting rules are manually created to suit the specific design of this application. For example, they can be designed to sort by signal energy from highest to lowest to ensure that the most important fault features are prioritized and processed. This sorting operation can more effectively extract and utilize key information from fault signals, providing strong support for subsequent time-frequency feature map conversion and the establishment of fault identification models.

[0082] Exemplarily, the present application uses an improved empirical wavelet transform to decompose the fault signal to be decomposed, sorts the modal functions obtained by decomposing a single signal in descending order according to the center frequency (i.e., the first sorting in the present application), arranges all decomposed signals in the order of three-phase voltage, three-phase current, zero-sequence voltage and zero-sequence current (i.e., the second sorting in the present application), and generates a time-frequency matrix (i.e., the first time-frequency matrix).

[0083] It should be noted that the empirical wavelet transform applies the idea of ​​empirical mode decomposition to the wavelet transform, and uses the compact support framework of the wavelet transform to provide the empirical mode decomposition with a more complete and reliable mathematical theoretical basis, reduce the computational complexity, and overcome the modal aliasing problem of EMD in component extraction.

[0084] Specifically, the steps of the first decomposition operation may be as follows:

[0085] Step 1: Complete frequency band division of the Fourier spectrum of the fault signal

[0086] (1) The time domain signal is transformed into the frequency domain to obtain the single-sided Fourier spectrum of the original fault signal f(t);

[0087] F(w)=|FFT[f(t)]|

[0088] (2) Divide F(w) into m segments, take the maximum value of each segment and its horizontal coordinate as the node and record it as A j , M is the data length of the discrete spectrum F(w);

[0089] (3) For all nodes A j Perform cubic spline interpolation to generate the envelope curve of the signal spectrum;

[0090] (4) The minimum value of the envelope curve is used as the spectrum frequency division boundary, and Ω is obtained. i =[w i-1 ,w i ],i=1,2,...,N,w i represents the center frequency of the i-th frequency band;

[0091] (5) w i Construct the scaling function and wavelet function group of the empirical wavelet transform, and perform EWT decomposition to obtain each EWT component. and the empirical wavelet function φ i (w) are:

[0092]

[0093] in:

[0094] τ i =γωi

[0095]

[0096] Among them, τi represents the width parameter related to wi, which is used to define the range of the scaling function and wavelet function, and β(x) represents an auxiliary function for smooth transition.

[0097] Step 2: Construct empirical wavelet transform. Similar to the method of constructing detail function and approximate function with wavelet transform, the empirical wavelet function φ i (t) and empirical scaling function Perform inner product operations with the original fault signal f(t) to obtain the detail function

[0098]

[0099] Among them, F -1 (.) is Fourier transform, i = 1, 2, ..., N;

[0100] Step 3: Reconstruct the original fault signal f(t). When decomposing the signal, the empirical wavelet method adds a bunch of white noise data with equal amplitude and opposite sign, thus constructing two sets of signals to be decomposed. The EWT decomposition process is then performed on each set of signals. This process is repeated multiple times, and the arithmetic mean of the intrinsic mode components during each repetition is calculated to reconstruct the signal.

[0101] (1) Add a pair of white noise sequences with opposite positive and negative values, a mean of 0 and a standard deviation of σ to the original fault signal f(t) i (t)(i=1,...,N), construct f in the i-th cycle i1 (t) and f i2 (t):

[0102]

[0103] Among them, noise i (t)~N(0,σ), obeys Gaussian distribution, f i1 (t) and f i2 (t) are two new signals to be decomposed;

[0104] (2) For f i1 (t) and f i2 (t) Perform empirical wavelet decomposition to obtain two sets of IMF components:

[0105]

[0106] IMF1 ij (t) is the signal f in the i-th decomposition operationi1 (t) The jth IMF component (j=1,…k) decomposed, IMF2 ij (t) Similarly, k is the maximum decomposition mode function;

[0107] (3) After repeating steps (1) and (2) in step 3 to add auxiliary white noise N times, the arithmetic mean of the IMF components in each cycle is calculated:

[0108]

[0109] Where j = 1, ..., K, IMF j (t) is the jth IMF component of the original signal f(t);

[0110] (4) The decomposition result of EWT of the original signal f(t) after adding auxiliary white noise is:

[0111]

[0112] in, represents convolution, They are and The Fourier transform of .

[0113] (4) Decomposed empirical mode function f k (t) is:

[0114]

[0115] It should be noted that performing a first decomposition operation on the second fault recording signal waveform to obtain a first time-frequency matrix can clearly show the energy distribution of the fault signal at different times and frequencies, providing intuitive and rich information for subsequent feature extraction and fault identification. Through in-depth analysis of the first time-frequency matrix, key features in the fault signal, such as frequency offset, energy mutation, etc., can be accurately captured. These features are crucial for determining the type of fault and locating the fault location. In addition, the generation of the first time-frequency matrix also provides effective input data for subsequent machine learning algorithms, enabling the fault identification model to more accurately learn and identify fault features, thereby improving the accuracy and reliability of fault identification. Therefore, performing a first decomposition operation on the second fault recording signal waveform to obtain a first time-frequency matrix is ​​one of the important steps in the fault identification method in the embodiment of the present application.

[0116] S103, converting the first time-frequency matrix into a second time-frequency feature map;

[0117] It should be noted that the purpose of converting the first time-frequency matrix into the second time-frequency feature graph is to further process and utilize the fault information contained in the first time-frequency matrix. This embodiment of the present application proposes a step of converting the first time-frequency matrix into the second time-frequency feature graph. This conversion process aims to convert matrix data into an image format, thereby more intuitively displaying the time-frequency characteristics of the fault signal and providing input for subsequent image processing and pattern recognition algorithms.

[0118] In an optional embodiment, the conversion of the first time-frequency matrix into the second time-frequency feature map can be achieved through image processing technology, such as using grayscale images, color images, or pseudo-color images to represent the data in the first time-frequency matrix. Among them, the grayscale image represents the strength of the data with different grayscale levels, and the color image or pseudo-color image represents the difference in data with different colors. In this way, the time-frequency characteristics of the fault signal can be displayed in a more intuitive and easy-to-understand manner, providing strong support for subsequent pattern recognition.

[0119] In an optional embodiment, during the process of converting the first time-frequency matrix into the second time-frequency feature map, the image may be enhanced, filtered, or segmented according to actual needs to further improve the image quality and the accuracy of feature extraction. For example, techniques such as histogram equalization and Laplace filtering may be used to enhance the contrast and details of the image, or techniques such as threshold segmentation and edge detection may be used to extract key features in the image. These processing techniques can be selected and optimized based on specific application scenarios and needs to achieve optimal image processing and feature extraction effects.

[0120] It should be noted that in order to more specifically convert the first time-frequency matrix into the second time-frequency feature map, the present application pixelates the first time-frequency matrix, that is, each element in the first time-frequency matrix corresponds to a pixel in the image, and the color or grayscale value of each pixel is determined according to the value of the corresponding element in the first time-frequency matrix. In this way, the fault information in the first time-frequency matrix can be intuitively displayed in the form of an image, which facilitates subsequent pattern recognition and fault identification.

[0121] In the embodiment of the present application, converting the first time-frequency matrix into the second time-frequency feature map includes:

[0122] Performing quantization encoding on the first time-frequency matrix;

[0123] Quantization coding is used to convert the first time-frequency matrix into a pixel matrix;

[0124] The quantization coding converts the first time-frequency matrix into a pixel matrix within the target pixel interval range through a random rounding operation to obtain a second time-frequency feature map.

[0125] In an optional embodiment, there are many operations for quantization coding. For example, different quantization strategies such as uniform quantization, non-uniform quantization or adaptive quantization can be used to adapt to the characteristics and identification requirements of different fault signals. Uniform quantization divides the numerical range in the first time-frequency matrix into several quantization intervals evenly, and each quantization interval corresponds to a pixel value. This method is simple to implement, but may not fully reflect the detailed features in the fault signal. Non-uniform quantization divides the numerical range into quantization intervals of different sizes according to the statistical characteristics of the fault signal to better capture the important features in the fault signal. Adaptive quantization dynamically adjusts the division of quantization intervals according to the changes in the fault signal to achieve more accurate quantization coding.

[0126] In the embodiment of the present application, in order to control the amount of calculation and improve the real-time performance of fault identification, the present application preferably adopts the following encoding method:

[0127] The time-frequency signal matrix obtained by the improved empirical wavelet transform decomposition is converted into a pixel matrix with image properties, and the round(.) function is used to round the time-frequency matrix to the pixel range of [0,255].

[0128]

[0129] Where P(m,n) represents the pixel value of the mth row and nth column of the pixel matrix, x i represents the value of the i-th data point, x max and x min Indicates the maximum and minimum sample values. Signal-to-image processing methods can convert a time-frequency data matrix into a pixel matrix with image properties without any parameters or human intervention, thereby enhancing information that is not visible in the original signal.

[0130] It should be noted that converting the first time-frequency matrix into the second time-frequency feature diagram can display complex time-frequency data in an intuitive and easy-to-understand image form, greatly improving the readability and usability of fault information. In this way, it is not only convenient for technical personnel to conduct in-depth analysis and understanding of the fault signal, but also provides more effective input data for subsequent automated processing and pattern recognition algorithms. In addition, this conversion process also enhances the hidden information in the original signal, making the fault characteristics more prominent, thereby improving the accuracy and efficiency of fault identification. Therefore, converting the first time-frequency matrix into the second time-frequency feature diagram is one of the key steps of the fault identification method in the embodiment of the present application, which is of great significance for achieving efficient fault identification.

[0131] S104, performing a sample expansion operation on the second time-frequency feature graph, and using the second time-frequency feature graph after sample expansion and the original second time-frequency feature graph as a first fault sample set;

[0132] In an optional embodiment, sample expansion is performed because, in practical applications, fault signals are often complex and diverse. The number of directly acquired fault samples may be limited, making it difficult to meet the large number of training samples required by algorithms such as deep learning. Therefore, by performing a sample expansion operation on the second time-frequency feature map, the diversity and number of fault samples can be increased, thereby improving the generalization ability and accuracy of the fault identification model. Sample expansion operations can be performed using a variety of methods, such as rotation, flipping, translation, scaling, and adding noise. These methods can effectively simulate the changes in fault signals under different conditions, thereby generating more representative fault samples.

[0133] In order to actually deal with the problems that may be encountered in the application scenarios of the method proposed in this application, this application designs a sample expansion operation.

[0134] In the embodiment of the present application, performing a sample expansion operation on the second time-frequency feature map includes:

[0135] Determine a fault type label for generating a first fault recording signal waveform of the target distribution network, and perform a separate encoding operation on the fault type label;

[0136] The first sample expansion model is trained using the fault type label after the separate encoding operation and the second time-frequency feature map representing the current system working condition as the conditional value and the real sample respectively;

[0137] The output of the first sample expansion model is the second time-frequency feature graph after sample expansion, and the acquired second time-frequency feature graph during distribution network fault is the original second time-frequency feature graph.

[0138] For example, data labels are added to the time-frequency matrix of fault features under different working conditions, and the fault labels are encoded at the same time, and a conditional generative adversarial network is used to expand the time-frequency feature map of the fault data;

[0139] Step 1: Perform one-hot encoding on different fault conditions;

[0140] The number of distribution network fault samples is relatively small, and the number of samples of different fault types is severely unbalanced. This causes deep learning models to overfit due to insufficient samples. Therefore, a conditional generative adversarial network (CGAN) is used to expand sample information. CGAN introduces conditional information into the inputs of the generative and discriminative models, enabling the generative model to learn the mapping relationship between the sample probability distribution under the corresponding conditional information and generate samples that meet the specified conditions. The fault type label is a discrete value with classification characteristics, so it needs to be one-hot encoded. The fault encoding is as follows:

[0141]

[0142]

[0143] In step ②, after fault coding, the fault type label and the time-frequency characteristic diagram representing the current system operating conditions are combined to expand the distribution network fault sample data.

[0144] CGAN is a game between the maximum and minimum values ​​of the generative model and the discriminative model with conditional constraints. It mines the deep relationships hidden in the data and makes the deep data infinitely close to the real data. The objective function of CGAN overall training is:

[0145]

[0146] Where D represents the mapping relationship of the discriminant model, G represents the mapping relationship of the generative model, V(D,G) represents the objective function of the network, and E is the expectation of the corresponding distribution. The model reaches its optimal state when it cannot correctly distinguish whether the input data is real data or generated data.

[0147] The generation of fault feature maps during distribution network faults is a regression problem. The least squares loss function is used to improve the image quality. The specific calculation formula is as follows:

[0148]

[0149] In the formula, a represents the label of the fake data, b represents the label of the real data, and c represents the value that the generator expects the discriminator to believe in the data. Dropout technology is used to prevent overfitting of the network, and BN technology is used to alleviate the problems of gradient vanishing and gradient exploding.

[0150] In step ③, the time-frequency feature map of the expanded fault data and the time-frequency feature map of the original fault are fused to form a new fault feature sample set (i.e., the first fault sample set).

[0151] It should be noted that performing a sample expansion operation on the second time-frequency feature graph and using the second time-frequency feature graph after sample expansion and the original second time-frequency feature graph as the first fault sample set can significantly increase the number and diversity of fault samples, which is crucial for training deep learning models. Deep learning models usually require a large number of training samples to learn the inherent laws and characteristics of the data in order to improve the generalization ability and accuracy of the model. Through the sample expansion operation, the present application can generate more representative fault samples, which can cover more fault types and working conditions, thereby making the trained fault identification model more robust and reliable. In addition, the expanded fault samples and the original fault samples are combined to form a new fault feature sample set, which can provide richer data support for subsequent model training and testing, and further improve the accuracy and efficiency of fault identification. Therefore, performing a sample expansion operation on the second time-frequency feature graph is an important part of the fault identification method in the embodiment of the present application, and has positive significance for achieving efficient fault identification.

[0152] S105, establishing a first fault identification model, and using the first fault sample set as a sample set of the first fault identification model for model training;

[0153] In an optional embodiment, the establishment of the first fault identification model may include selecting a suitable deep learning framework and algorithm, such as a convolutional neural network (CNN), a recurrent neural network (RNN) or a generative adversarial network (GAN), etc., and constructing a model structure suitable for distribution network fault identification according to the specific needs and characteristics of fault identification. The design of the model structure should fully consider the time-frequency characteristics and complexity of the fault signal, as well as factors such as the diversity and quantity of samples. For example, multi-layer convolutional layers and pooling layers can be used to extract local features and global features in the fault signal, and fully connected layers can be used for classification and identification. At the same time, in order to enhance the generalization ability and robustness of the model, regularization, Dropout and other technologies can also be introduced to prevent overfitting.

[0154] In an optional embodiment, during model training, strategies such as cross-validation and early stopping are employed to optimize model parameters and performance. Furthermore, techniques such as transfer learning can be utilized to accelerate model training and convergence, improving model training efficiency and accuracy. The first fault sample set is used as the sample set for the first fault identification model for model training. Through continuous iteration and optimization, the model learns the key features and patterns in the fault signal, enabling accurate identification and location of different types of faults.

[0155] In the embodiment of the present application, a convolutional neural network is selected to establish a first fault identification model. Establishing the first fault identification model includes:

[0156] Compressing the first fault sample set to obtain a feature image of a target size;

[0157] The feature image of the target size is input into the initial untrained first fault recognition model;

[0158] The first fault identification model includes an input layer, two sets of alternating convolutional layers and pooling layers, a fully connected layer, and an output layer.

[0159] It should be noted that establishing a first fault identification model and using the first fault sample set as the sample set of the first fault identification model for model training can make full use of the rich data in the first fault sample set, including the original fault samples and the expanded fault samples. These data cover a variety of fault types and working conditions, providing comprehensive support for model training. Through training, the first fault identification model can learn the deep features and laws in the fault signal, thereby achieving accurate classification and identification of different types of faults. In addition, the use of deep learning algorithms such as convolutional neural networks can automatically extract features from fault signals, avoiding the tedious process of manually designing features in traditional methods, and improving the efficiency and accuracy of fault identification. Therefore, establishing a first fault identification model and performing model training is one of the core steps of the fault identification method in the embodiment of the present application, which is of great significance for achieving efficient fault identification.

[0160] S106 , performing target distribution network fault identification according to the trained first fault identification model.

[0161] In an optional embodiment, when it is necessary to identify the fault of the target distribution network, the fault signal of the target distribution network is first obtained, and then the fault signal is processed according to the above-mentioned method steps to obtain the corresponding second time-frequency feature map. Then, the second time-frequency feature map is input into the first fault identification model after training. The model will automatically extract features and perform classification, and finally output the fault identification result. The fault identification result may include information such as the fault type and fault location, which is of great guiding significance for subsequent fault handling and maintenance. In this way, the distribution network fault identification method provided in the embodiment of the present application can realize the rapid and accurate identification of the faults of the target distribution network, and provide a strong guarantee for the safe and stable operation of the power system.

[0162] In an optional embodiment, identifying a target distribution network fault according to the trained first fault identification model includes:

[0163] The first fault identification model also includes a classifier connected to the output layer;

[0164] The classifier is used to perform fault identification and classification on the output of the first fault identification model;

[0165] The process of building a classifier includes:

[0166] Calculating the mode of the first fault sample set, and calculating the similarity between the mode and each first fault sample;

[0167] Determining a weight of each first fault sample according to the similarity;

[0168] Construct a classifier based on the first fault sample set after weight determination.

[0169] For example, the extracted feature vector is input into the instance-weighted hidden naive Bayes classifier to complete the distribution network fault type identification. The instance-weighted hidden Bayes classifier algorithm introduces latent variables to enhance the model y i (i=1, 2 ,...,n) to capture the potential features and improve the accuracy of classification. The algorithm steps are as follows:

[0170] (1) Input the training set of fault characteristic signals of the first fault sample set of the distribution network and calculate the mode z of the training set;

[0171] (2) Calculate the mode z of the training set and each feature instance y i The similarity between s(z,y i ), and then calculate each training instance y i The weight w i (i=1,2,...,n);

[0172]

[0173] Then the weight of the i-th training instance is:

[0174] w i =1+s(x,y i )

[0175] where δ(α i (x),α i (y)) is a binary function, when α i (x) and α i The function value is 1 when (y) is equal, otherwise it is 0;

[0176] (3) Calculate the probability of each fault category, that is, the prior probability P(c);

[0177]

[0178] (4) Calculate the sample α to be judged under the premise of belonging to each fault category i In the attribute value c i The conditional probability, that is, P(α i|c);

[0179]

[0180] (5) Calculate the sample α to be judged i The probability P of belonging to each category j =P(α j |c i )j=1,2,...,m;

[0181] (6) Select the category with the largest probability as the category of the sample to be judged, that is, P τ =max{P1,P2,...,P m}, the sample to be judged is judged as the rth category;

[0182] (7) Output the identification result of the fault signal;

[0183] (8) The information push module pushes the identified fault type to the administrator of the corresponding line for processing.

[0184] It should be noted that, through the above steps, the distribution network fault identification method provided by the embodiment of the present application can make full use of sample data, especially by increasing the number and diversity of fault samples through sample expansion operations, providing richer and more comprehensive data support for the training of deep learning models. This not only improves the generalization ability and accuracy of the model, but also enables the model to better adapt to different types of faults and working conditions. At the same time, the use of advanced deep learning algorithms such as convolutional neural networks can automatically extract deep features and laws in fault signals, avoiding the tedious process of manually designing features in traditional methods. In addition, by introducing advanced technologies such as instance-weighted hidden naive Bayes classifiers, the accuracy and efficiency of fault identification are further improved. Therefore, the distribution network fault identification method provided by the embodiment of the present application has broad application prospects and important practical significance, and can provide a strong guarantee for the safe and stable operation of the power system.

[0185] In summary, the present application proposes a method for identifying distribution network faults, which includes obtaining a first fault recording signal waveform of a target distribution network, and performing a first preprocessing on the fault recording signal waveform to obtain a second fault recording signal waveform; performing a first decomposition operation on the second fault recording signal waveform to obtain a first time-frequency matrix; converting the first time-frequency matrix into a second time-frequency feature graph; performing a sample expansion operation on the second time-frequency feature graph, and using the second time-frequency feature graph after sample expansion and the original second time-frequency feature graph as a first fault sample set; establishing a first fault identification model, and using the first fault sample set as a sample set of the first fault identification model for model training; and identifying the target distribution network fault based on the trained first fault identification model. The present application combines signal processing technology with quantization coding to analyze distribution network faults from a two-dimensional image perspective, thereby enhancing unreliable information in the fault signal. At the same time, it proposes to use a generative adversarial network to expand the distribution network fault sample data from an image expansion perspective, thereby avoiding the problem of overfitting of the fault identification model due to insufficient training sample data, and improving the system's ability to identify different faults.

[0186] Example 2: In a preferred embodiment, the specific steps of establishing the first fault identification model may be as follows:

[0187] Compress the new fault feature sample set (i.e., the first fault sample set) into a 32*32 image as the input of the convolutional neural network;

[0188] Establishing a convolutional neural network for the first fault identification parameter for training, using the compressed image as the network input for training, and extracting the output of the fully connected layer of the network as the characteristic phasor of the distribution line fault;

[0189] A convolutional neural network consists of an input layer, a convolutional layer, an output layer, and a fully connected layer. It extracts features from data by alternating between convolutional and pooling layers. CNNs are used to mine signals from images that can characterize the fault.

[0190] (1) Input layer: compress the pixel matrix into 32*32 as the input of the network;

[0191] (2) Convolutional layer C1: When extracting graphic features, the convolution layer needs to set the size and number of convolution kernels in advance, set the moving step to 1, obtain 32 feature maps, and fill pixels around the original image to prevent the loss of image edge feature information;

[0192] (3) Pooling layer S1: Find the maximum value of the 3*3 area of ​​the feature map obtained by the C1 layer, set the moving step size to 2, and use ReLU as the activation function through weighting and bias to map the 32 32*32 feature maps extracted by the C1 layer into 32 15*15 feature maps;

[0193] (4) Convolutional layer C2: extracts the features of the output image of pooling layer S1 through 1024 convolution kernels, and obtains 32 15*15 feature images;

[0194] (5) Pooling layer S2 and pooling layer S3 use maximum pooling to reduce the dimension of feature vector.

[0195] (6) Fully connected layer fc1: The fc1 layer stretches and expands the vector obtained by the pooling layer S3 into a one-dimensional vector;

[0196] (7) Output layer: outputs the expanded fault sample feature vector.

[0197] It should be noted that these feature vectors not only contain information about the original fault samples but also incorporate the diversity of the expanded samples, thereby enhancing the model's recognition capabilities. The output layer design must ensure that these feature vectors can be accurately and efficiently transmitted, providing a solid foundation for subsequent classification and recognition tasks.

[0198] Embodiment 3: This embodiment further provides a distribution network fault identification system, including:

[0199] A data acquisition and processing module is used to acquire a first fault recording signal waveform of the target distribution network, and perform a first preprocessing on the fault recording signal waveform to obtain a second fault recording signal waveform;

[0200] a matrix acquisition module, configured to perform a first decomposition operation on the second fault recording signal waveform to obtain a first time-frequency matrix;

[0201] A feature map acquisition module, configured to convert the first time-frequency matrix into a second time-frequency feature map;

[0202] An expansion module, configured to perform a sample expansion operation on the second time-frequency feature graph, and use the second time-frequency feature graph after sample expansion and the original second time-frequency feature graph as a first fault sample set;

[0203] A model building module is used to build a first fault identification model and use the first fault sample set as a sample set of the first fault identification model for model training;

[0204] The identification module is used to identify the target distribution network fault according to the first fault identification model after training.

[0205] The above-mentioned unit modules can be embedded in or independent of the processor in the electronic device in the form of hardware, or can be stored in the memory of the electronic device in the form of software, so that the processor can call and execute the corresponding operations of the above-mentioned modules.

[0206] This embodiment also provides an electronic device, which may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 2 As shown. The electronic device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the electronic device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for identifying a distribution network fault is implemented. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic device can be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the electronic device, or an external keyboard, touchpad or mouse.

[0207] This embodiment further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the following steps are implemented:

[0208] Acquire a first fault recording signal waveform of the target distribution network, and perform a first preprocessing on the fault recording signal waveform to obtain a second fault recording signal waveform;

[0209] Performing a first decomposition operation on the second fault recording signal waveform to obtain a first time-frequency matrix;

[0210] Converting the first time-frequency matrix into a second time-frequency feature map;

[0211] Performing a sample expansion operation on the second time-frequency feature graph, and using the second time-frequency feature graph after sample expansion and the original second time-frequency feature graph as a first fault sample set;

[0212] Establishing a first fault identification model, and using the first fault sample set as a sample set of the first fault identification model for model training;

[0213] The target distribution network fault is identified based on the first fault identification model after training.

[0214] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, and all of these should be included in the scope of the claims of the present application.

[0215] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiment of the present application can be implemented in various computer languages.

[0216] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0217] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0218] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0219] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0220] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A distribution network fault identification method, characterized in that: include: Acquire a first fault recording signal waveform of a target distribution network, and perform a first preprocessing on the fault recording signal waveform to obtain a second fault recording signal waveform; Performing a first decomposition operation on the second fault recording signal waveform to obtain a first time-frequency matrix; Converting the first time-frequency matrix into a second time-frequency feature map; Performing a sample expansion operation on the second time-frequency feature graph, and using the second time-frequency feature graph after sample expansion and the original second time-frequency feature graph as a first fault sample set; Establishing a first fault identification model, and using the first fault sample set as a sample set of the first fault identification model for model training; The target distribution network fault is identified based on the first fault identification model after training.

2. A distribution network fault identification method according to claim 1, characterized in that: The performing a first decomposition operation on the second fault recording signal waveform includes: performing a first frequency band division on the waveform of the second fault recording signal to obtain a plurality of frequency bands; Establishing a plurality of filters for each frequency band, and obtaining a plurality of fault filtering signals according to the plurality of filters; A target fault filter signal waveform modal function is obtained according to the plurality of fault filter signals.

3. A distribution network fault identification method according to claim 2, characterized in that: The performing a first decomposition operation on the second fault recording signal waveform further includes: Performing a first sorting of the target fault filter signal waveform modal functions according to center frequencies; and performing a second sorting on all target fault filter signal waveform modal functions; A first time-frequency matrix is ​​obtained according to the first sorting and the second sorting results.

4. A distribution network fault identification method according to claim 3, characterized in that: The converting the first time-frequency matrix into a second time-frequency feature map includes: Performing quantization encoding on the first time-frequency matrix; The quantization coding is used to convert the first time-frequency matrix into a pixel matrix; The quantization coding converts the first time-frequency matrix into a pixel matrix within a target pixel interval range through a random rounding operation to obtain a second time-frequency feature map.

5. A distribution network fault identification method as claimed in claim 4, characterized in that: The performing a sample expansion operation on the second time-frequency feature map includes: Determining a fault type label for generating a first fault recording signal waveform of a target distribution network, and performing a separate encoding operation on the fault type label; The first sample expansion model is trained using the fault type label after the separate encoding operation and the second time-frequency feature map representing the current system working condition as the conditional value and the real sample respectively; The output of the first sample expansion model is the second time-frequency feature graph after sample expansion, and the acquired second time-frequency feature graph during distribution network failure is the original second time-frequency feature graph.

6. A distribution network fault identification method according to claim 5, characterized in that: The establishing of the first fault identification model comprises: Compressing the first fault sample set to obtain a feature image of a target size; Taking the characteristic image of the target size as input, and inputting it into an initially untrained first fault recognition model; The first fault identification model includes an input layer, two sets of alternating convolutional layers and pooling layers, a fully connected layer, and an output layer.

7. A distribution network fault identification method according to claim 6, characterized in that: The target distribution network fault identification according to the trained first fault identification model includes: The first fault identification model further includes a classifier connected to the output layer; The classifier is used to perform fault identification and classification on the output of the first fault identification model; The process of establishing the classifier includes: Calculating the mode of the first fault sample set, and calculating the similarity between the mode and each first fault sample; determining a weight of each first fault sample according to the similarity; Construct a classifier based on the first fault sample set after weight determination.

8. A distribution network fault identification system, applying the method according to any one of claims 1 to 7, characterized in that: include: A data acquisition and processing module is used to acquire a first fault recording signal waveform of a target distribution network, and perform a first preprocessing on the fault recording signal waveform to obtain a second fault recording signal waveform; a matrix acquisition module, configured to perform a first decomposition operation on the second fault recording signal waveform to obtain a first time-frequency matrix; A feature map acquisition module, configured to convert the first time-frequency matrix into a second time-frequency feature map; an expansion module, configured to perform a sample expansion operation on the second time-frequency characteristic graph, and use the second time-frequency characteristic graph after sample expansion and the original second time-frequency characteristic graph as a first fault sample set; A model building module, configured to build a first fault identification model and use the first fault sample set as a sample set of the first fault identification model for model training; The identification module is used to identify the target distribution network fault according to the first fault identification model after training.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a distribution network fault identification method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a distribution network fault identification method according to any one of claims 1 to 7 are implemented.

Images