Distribution network line fault positioning method, device, equipment and medium

Abstract

The invention discloses a distribution network line fault positioning method, device and equipment and a medium, and the method comprises the steps: carrying out the full mining of fault features through integrating multi-dimensional feature vectors of transient traveling wave signals, simulating the distribution and features of real data through GAN, enhancing scarce high-resistance fault data, expanding training samples, and achieving the fault positioning of a distribution network line. And the model can learn a wider fault feature mode. Afterwards, a long short-term memory (LSTM) network is adopted to train a distribution network line fault positioning model, high-frequency characteristics of weak signals generated by rare faults such as high-resistance grounding and intermittent arc can be effectively captured, the problem of high-resistance fault detection of the distribution network line is solved, and the accuracy of distribution network line fault positioning is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid faults, and in particular to a method, device, equipment and medium for locating a distribution line fault. Background Art

[0002] Related technologies for fault location in distribution network overhead lines rely on modeling methods driven by historical data. However, due to the sporadic nature of fault events and the limited coverage of monitoring equipment, historical datasets lack samples for complex scenarios such as high-resistance ground faults and indirect arc faults. This imbalance in data distribution causes the trained models to be heavily biased towards common fault types, while significantly lacking generalization capabilities for marginal fault scenarios, ultimately leading to a significant increase in fault location errors. Summary of the Invention

[0003] The present invention provides a distribution line fault location method, device, electronic device and medium to solve the technical problem that samples of complex scenarios such as high-resistance grounding faults and indirect arc faults are scarce in historical data sets, causing the trained model to be seriously biased towards common fault types, resulting in a significant increase in fault location errors.

[0004] In a first aspect, a method for locating a distribution line fault is provided, comprising:

[0005] Based on historical fault recording data of the distribution line, determining a first transient traveling wave signal and a second transient traveling wave signal at the beginning and end of the line, and performing power frequency interference removal and noise reduction processing on the first transient traveling wave signal and the second transient traveling wave signal to obtain a first reconstructed signal and a second reconstructed signal;

[0006] Extracting eigenvalues ​​of the first reconstructed signal and the second reconstructed signal, normalizing the extracted eigenvalues ​​and combining them to obtain a eigenvector;

[0007] Based on the feature vector and preset condition information, a synthetic feature vector is generated through a conditional generative adversarial network;

[0008] A long short-term memory model is constructed and trained based on the eigenvectors and synthetic eigenvectors to obtain a distribution line fault location model.

[0009] The real-time signal of the distribution line is obtained and input into the distribution line fault location model to obtain the fault location of the distribution line.

[0010] In a second aspect, a distribution line fault location device is provided, comprising:

[0011] a determination module for determining, based on historical fault recording data of the distribution line, a first transient traveling wave signal and a second transient traveling wave signal at the beginning and end of the line; a first acquisition module for performing power frequency interference removal and noise reduction processing on the first transient traveling wave signal and the second transient traveling wave signal to obtain a first reconstructed signal and a second reconstructed signal;

[0012] A feature extraction module is used to extract feature values ​​from the first reconstructed signal and the second reconstructed signal, and to normalize the extracted feature values ​​and combine them to obtain a feature vector;

[0013] A first generation module is used to generate a synthetic feature vector based on the feature vector and preset condition information through a conditional generative adversarial network;

[0014] The second generation module is used to build a long short-term memory model, train the long short-term memory model based on the feature vector and the synthetic feature vector, and obtain a distribution line fault location model;

[0015] The second acquisition module is used to obtain the real-time signal of the distribution line; the third generation module is used to input the real-time signal into the distribution line fault location model to obtain the fault location of the distribution line.

[0016] In a third aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned distribution line fault location method when executing the computer program.

[0017] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned distribution line fault location method are implemented.

[0018] The solution implemented by the aforementioned distribution line fault location method, device, electronic device, and storage medium fully exploits fault characteristics by integrating the multidimensional feature vectors of transient traveling wave signals. Generative Adversarial Networks (GANs) are then used to simulate the distribution and characteristics of real-world data, enhancing scarce high-resistance fault data and expanding the training sample base, enabling the model to learn a wider range of fault characteristic patterns. Subsequently, a long short-term memory (LSTM) network is employed to train the distribution line fault location model. This effectively captures the high-frequency characteristics of weak signals generated by rare faults such as high-resistance grounding and intermittent arcing, addressing the challenge of detecting high-resistance faults in distribution network lines and significantly improving the accuracy of distribution line fault location. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 1 is a flow chart of a method for locating a distribution line fault in one embodiment of the present invention;

[0021] Figure 2 1 is a flow chart of a method for locating faults in a multi-source data distribution network overhead line based on LSTM-GAN in a specific embodiment of the present invention;

[0022] Figure 3 1 is a schematic structural diagram of a distribution line fault locating device in one embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. It should be understood that the drawings in the present invention are only for the purpose of illustration and description and are not used to limit the scope of protection of the present invention.

[0024] In addition, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in the present invention illustrate operations implemented according to some embodiments of the present invention. It should be understood that the operations in the flowcharts may be implemented out of sequence, and steps that do not have a logical contextual relationship may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the present disclosure, may add one or more other operations to the flowcharts, or may remove one or more operations from the flowcharts.

[0025] In addition, the embodiments described in the present invention are only some of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0026] It should be noted that the term "comprising" will be used in the embodiments of the present invention to indicate the presence of the features subsequently claimed, but does not preclude the addition of other features. It should also be noted that similar reference numerals and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined or explained in subsequent figures.

[0027] The following is a detailed description of this case with reference to the relevant drawings in the specification.

[0028] In the embodiments of this specification, the current distribution network overhead line fault location technology mainly adopts a historical data-driven modeling method. However, in actual applications, it faces the following challenges: Sample distribution imbalance problem: Due to the sporadic nature of fault events and insufficient coverage of monitoring equipment, complex fault samples such as high-resistance grounding (>500Ω) and indirect arcs account for less than 5%, resulting in the model training set being severely biased towards common fault types such as low-resistance short circuits (accounting for 85%). This data skew significantly reduces the model's recognition accuracy for edge scenarios.

[0029] Furthermore, related technologies rely on static line parameters (such as wave velocity, inductance per unit length, and capacitance). Actual line parameters vary dynamically with ambient temperature, humidity, and aging. Calibration requires downtime every 6-12 months, making it impossible to adapt to parameter drift in real time, leading to cumulative positioning errors.

[0030] Furthermore, traveling wave location technology faces the following challenges in multi-branch (≥3 branches) and ring network structures: aliasing of reflected waves and uneven signal attenuation make it difficult to accurately distinguish between mainline and branch interference. Traditional methods rely on two key simplifying assumptions: constant wave velocity and ideal conditions with no branch reflections. These assumptions deviate from the dynamic propagation process, causing the location results to deviate from the actual fault point.

[0031] Furthermore, traveling wave location technology has limited ability to capture weak transient signals on distribution lines (such as high-resistance fault currents <1A). Traditional sensors (such as electromagnetic CTs) experience severe signal attenuation in low signal-to-noise ratio (SNR <10dB) environments, and their threshold triggering mechanisms easily filter out weak wave heads, resulting in a high-resistance fault miss rate exceeding 50%. These limitations collectively limit the accuracy and reliability of precise location of distribution overhead lines.

[0032] To address these issues, this application proposes a multi-source data-based LSTM-GAN-based method for locating overhead line faults in distribution networks. By integrating the multidimensional feature vectors of transient traveling wave signals, this method fully explores fault characteristics. GANs are then used to simulate the distribution and characteristics of real data, enhancing scarce high-resistance fault data and expanding training samples, enabling the model to learn a wider range of fault characteristic patterns. Subsequently, a long short-term memory (LSTM) network is used to train a distribution line fault location model. This method effectively captures the high-frequency characteristics of weak signals generated by rare faults such as high-resistance grounding and intermittent arcing, addressing the challenge of detecting high-resistance faults in distribution network lines and significantly improving the accuracy of distribution line fault location.

[0033] See also Figure 1 This embodiment of the present invention provides a method for locating a distribution line fault, which specifically includes the following steps:

[0034] S10: Based on the historical fault recording data of the distribution line, determine the first transient traveling wave signal and the second transient traveling wave signal at the beginning and end of the line, and remove the power frequency interference and noise reduction processing on the first transient traveling wave signal and the second transient traveling wave signal to obtain a first reconstructed signal and a second reconstructed signal.

[0035] It is understandable that the execution subject of the present invention may be a distribution line fault location device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0036] In this step, the fault event data recorded by the distribution automation system (SCADA), fault recording device or traveling wave ranging device is obtained, wherein the historical fault recording data contains the complete transient process before and after the fault occurs (such as 1 cycle before the fault + 10 cycles after the fault). From the historical fault recording data of the distribution line, the voltage / current transient traveling wave signals at the beginning and end of the line are synchronously extracted and recorded as the first transient traveling wave signal and the second transient traveling wave signal. Among them, the strong fundamental component (50Hz / 60Hz) and its harmonics (100Hz, 150Hz, etc.) are dominant, which may mask the transient characteristics of the fault. Therefore, an adaptive filtering algorithm is used to filter out the 50Hz / 60Hz power frequency and its harmonic interference. In addition, the signal includes Gaussian white noise (sensor noise), pulse noise (switching operation) and high-frequency oscillation (line capacitance effect). Therefore, combined with wavelet threshold denoising, high-frequency noise and transient interference are eliminated, and finally high-quality first reconstructed signal and second reconstructed signal are output.

[0037] Among them, for the first transient traveling wave signal x A (t) and the second transient traveling wave signal x B (t) Perform power frequency interference removal and noise reduction processing, specifically including:

[0038] (1) Removal of power frequency interference: Use Butterworth high-pass filter (cut-off frequency f c =1KHz), filter out the power frequency component.

[0039] Difference equation implementation:

[0040]

[0041] Among them, the coefficient {b k}、{a k}Generated by filter design tool.

[0042] (2) Noise reduction: Use wavelet threshold denoising, taking Symlet8 wavelet as an example:

[0043] Decompose the wavelet into 5 layers to obtain detail coefficients d1-d5 and approximate coefficient a5;

[0044] Apply software thresholding to the high frequency detail coefficients (d1-d3) Where σ is the noise standard deviation to suppress noise.

[0045] Finally, wavelet reconstruction obtains the first reconstructed signal and the second reconstructed signal

[0046] In actual application scenarios, sensors are installed at both ends of the distribution line (end A and end B), and equipment configuration requirements are set (such as sampling rate ≥ 1MHz to ensure that nanosecond surge signals can be captured; time synchronization error is less than or equal to 1μs), and signals are collected synchronously at ends A and B.

[0047] Through the above method, the power frequency and noise are removed, the signal-to-noise ratio (SNR) is improved, and the accuracy of subsequent feature extraction is ensured.

[0048] S20: Extracting eigenvalues ​​from the first reconstructed signal and the second reconstructed signal, normalizing the extracted eigenvalues, and combining them to obtain a eigenvector.

[0049] In this step, key features are extracted from the first and second reconstructed signals, including time-domain features, frequency-domain features, and joint time-frequency features. Subsequently, the Min-Max normalization method is used to uniformly map feature values ​​of different dimensions to a specific interval, eliminating dimensional differences and accelerating model convergence. The normalized features are then concatenated in time sequence to form a multidimensional feature vector.

[0050] Through the above method, key features are extracted, multi-dimensional fault representation is integrated, and the ability to distinguish complex faults is enhanced, providing high-quality input features for subsequent machine learning algorithms, significantly improving the accuracy and reliability of fault diagnosis.

[0051] In one embodiment of the present application, a specific feature extraction scheme is provided. In S20, feature values ​​are extracted from the first reconstructed signal and the second reconstructed signal, and the extracted feature values ​​are normalized and then combined to obtain a feature vector. The scheme specifically includes the following steps S21-S25:

[0052] S21: Perform continuous wavelet transform on the first reconstructed signal and the second reconstructed signal respectively to generate a first time-frequency coefficient matrix and a second time-frequency coefficient matrix.

[0053] In this step, continuous wavelet transform (CWT) is a time-frequency analysis tool that convolves the signal with the wavelet basis function to obtain the energy distribution of the signal at different scales (frequency) and time. Haar wavelet (compact support in time domain, suitable for mutation detection) is used to perform continuous wavelet transform on the first reconstructed signal. i , calculate the convolution of the signal and the wavelet, and get the time series W(a i , b), and finally generate a two-dimensional time-frequency matrix W A (a i , b j )(size: ruler reading × number of time points), recorded as the first time-frequency coefficient matrix, where the horizontal axis (time axis) is aligned with the original signal time, marking the moment of the transient event (such as the arrival time of the wave head); the vertical axis (scale axis / frequency axis) is the frequency band corresponding to different scales a, high frequency (small scale) captures rapid mutations, and low frequency (large scale) reflects slow components; the coefficient amplitude |W(a i , b j )| represents the energy intensity of the signal at this time-frequency point. Further, the second reconstructed signal is transformed by continuous wavelet transform to obtain the matrix W B (a i , b j ), recorded as the second time-frequency coefficient matrix.

[0054] The formula for calculating the wavelet coefficients is:

[0055]

[0056] Where a is the size parameter (determines the frequency, the smaller a is, the higher the frequency is); b is the translation parameter (determines the time position); ψ(t) is the Haar wavelet basis function.

[0057] S22: Based on the preset size, extract multiple first modulus maximum points in the first time-frequency coefficient matrix and obtain the first arrival time of each modulus maximum point, and extract multiple second modulus maximum points in the second time-frequency coefficient matrix and obtain the second arrival time of each modulus maximum point.

[0058] In this step, if the amplitude of the wavelet coefficient at a certain point in the time-frequency coefficient matrix presents a local maximum value at both adjacent scales and adjacent times, then this point is called the modulus maximum point, which corresponds to the mutation point in the signal (such as the wave head of the power traveling wave, the starting point of the fault transient). The time coordinate b corresponding to the modulus maximum point is j , reflecting the specific moment when the signal characteristics (such as wave head) arrive at the sensor.

[0059] Specifically, taking the first time-frequency coefficient matrix as an example, Perform continuous wavelet transform to obtain the coefficient matrix W A (a i , b j ). Fix the scale a=2 to extract the coefficient W of the high frequency band A (2, b j Traverse the time axis b and mark it as the first modulus maximum point if the following conditions are met:

[0060] |W(2,b j )|>|W(2,b j-1 )|and|W(2,b j )|>|W(2,b j+1) ∣;

[0061] Afterwards, for each first modulus maximum point, record its time coordinate b j As the first arrival time.

[0062] Furthermore, the same operation is performed on the second time-frequency coefficient matrix to obtain a plurality of second modulus maximum points, and the second arrival time of each modulus maximum point is obtained.

[0063] S23: Determine characteristic values ​​of multiple target types based on the first reconstructed signal, the second reconstructed signal, multiple first arrival times of multiple first modulus maximum points, and multiple second arrival times of multiple second modulus maximum points.

[0064] Among them, the various target types include: traveling wave head time difference, transient surge amplitude characteristics, transient surge polarity characteristics, frequency domain energy ratio characteristics and main frequency wave velocity characteristics.

[0065] In this step, five categories of key feature quantities are extracted based on the first reconstructed signal, the second reconstructed signal, the multiple first arrival times of the multiple first modulus maximum points, and the multiple second arrival times of the multiple second modulus maximum points to achieve multi-dimensional feature expression of the transient traveling wave signal.

[0066] In one embodiment of the present application, a specific key feature extraction scheme is provided. In S23, based on the first reconstructed signal, the second reconstructed signal, the first arrival times of the first modulus maximum points, and the second arrival times of the second modulus maximum points, feature values ​​of multiple target types are determined, specifically including:

[0067] Obtaining a first initial traveling wave arrival time of a first modulus maximum point among the plurality of first modulus maximum points, and a second initial traveling wave arrival time of a first modulus maximum point among the plurality of second modulus maximum points;

[0068] Calculating the absolute value of the difference between the first initial traveling wave arrival time and the second initial traveling wave arrival time to obtain the traveling wave front time difference;

[0069] Obtaining a first sign of a wavelet coefficient at a first initial traveling wave arrival time and a second sign of a wavelet coefficient at a second initial traveling wave arrival time;

[0070] determining a first transient surge polarity at an arrival time of a first initial traveling wave based on the first symbol;

[0071] A second transient surge polarity of the second initial traveling wave arrival time is determined based on the second sign.

[0072] In this embodiment, key feature value extraction specifically includes: traveling wave head time difference calculation. Specifically, the first modulus maximum point refers to the first modulus maximum that appears on the time axis, corresponding to the arrival time of the initial traveling wave head, which directly reflects the exact moment when the initial traveling wave generated by the fault point arrives at the measurement end. In transient traveling wave analysis, the first initial traveling wave arrival time t1 corresponding to the first modulus maximum point of the initial traveling wave arriving at the A-end sensor and the second initial traveling wave arrival time t2 corresponding to the first modulus maximum point of the initial traveling wave arriving at the B-end sensor are obtained, and the traveling wave head time difference Δt = |t1-t2| is calculated to determine the electromagnetic wave propagation delay between the fault point and the measurement end.

[0073] Furthermore, key feature extraction also includes transient surge polarity (P) extraction. Transient surge polarity extraction refers to the process of quantitatively calibrating the sudden change direction of each traveling wave head detected in the signal. By analyzing the sign characteristics of the wavelet transform coefficients, key polarity feature information is provided for fault detection. Specifically, each wave head corresponds to a transient surge, and the wavelet coefficient is the inner product of the signal and the wavelet basis function. The sign of its actual part reflects the sudden change direction of the signal at a specific moment:

[0074] W(a, b)>0: The signal presents a positive mutation at this moment (such as a voltage surge);

[0075] W(a, b)<0: the signal presents a negative mutation (such as voltage drop) at this moment.

[0076] The traveling wave head is the starting mutation point of the transient signal, and its polarity directly reflects the initial voltage / current change direction caused by the fault or lightning strike:

[0077] P=+1: The wave head is a rising edge (such as a positive polarity surge caused by lightning);

[0078] P=-1: The wave head is a falling edge (such as a negative polarity surge caused by a short circuit).

[0079] During the transient analysis, the signs of the wavelet coefficients are checked for the first initial traveling wave arrival time t1 and the second initial traveling wave arrival time t2, respectively, to obtain the first transient surge polarity and the second transient surge polarity. Taking the initial traveling wave arrival time t1 as an example, extract W A The real part Re[W A (2, t1)], judgment symbol:

[0080]

[0081] For example, if Re[W A (2, t1)] = +0.25, then P = +1, positive surge; if Re [W A (2, t1)] = -0.18, then P = -1, negative surge.

[0082] In one embodiment of the present application, a specific key feature extraction scheme is provided. In S23, based on the first reconstructed signal, the second reconstructed signal, the first arrival times of the first modulus maximum points, and the second arrival times of the second modulus maximum points, feature values ​​of multiple target types are determined, which specifically includes:

[0083] Obtaining a first absolute amplitude corresponding to each first arrival time, and converting the first absolute amplitude into a first per-unit value to obtain a first transient surge amplitude;

[0084] A second absolute amplitude corresponding to each second arrival time is obtained, and the second absolute amplitude is converted into a second per-unit value to obtain a second transient surge amplitude.

[0085] In this embodiment, the key feature value extraction also includes: surge amplitude (A pu ) extraction. Specifically, the extraction of each surge amplitude is the process of converting the absolute amplitude of the traveling wave signal into a per unit value (Per Unit), which is used to uniformly quantify the surge intensity and eliminate the influence of the voltage level. Specifically, the absolute amplitude is the instantaneous maximum deviation value (peak value) of the traveling wave head on the voltage / current signal, reflecting the intensity of the surge. For each wave head, at the arrival time t of each wave head, k In the nearby time window (usually t k±0.1μs), determine the absolute maximum value of the reconstructed signal. Take end A as an example:

[0086]

[0087] Among them, t k is the arrival time of the kth wave head; Δt is the time window.

[0088] For example, if the wave head The peak value is 15.3kV, then A k =15.3kV.

[0089] Afterwards, A k Perform normalization processing, that is, convert the absolute amplitude into a per-unit value.

[0090] The conversion formula is:

[0091]

[0092] Where U base As the reference voltage, it is usually the rated voltage of the distribution line U base =220kV as the benchmark.

[0093] Optionally, if the initial amplitude of the fault point needs to be restored, the initial amplitude can be calculated by combining the traveling wave amplitude attenuation model with distance.

[0094] The attenuation model is:

[0095] A 0,pu =A pu e αx ;

[0096] Where x is the surge propagation distance; α is the attenuation coefficient, in Np / km, and α is determined by the line parameters:

[0097]

[0098] Where R is the resistance per unit length; G is the conductance; is the characteristic impedance.

[0099] In one embodiment of the present application, a specific key feature extraction scheme is provided. In S23, based on the first reconstructed signal, the second reconstructed signal, the first arrival times of the first modulus maximum points, and the second arrival times of the second modulus maximum points, feature values ​​of multiple target types are determined, which specifically includes:

[0100] Performing a fast Fourier transform on the first reconstructed signal to obtain a first spectrum;

[0101] Performing a fast Fourier transform on the second reconstructed signal to obtain a second spectrum;

[0102] Calculating a first high-frequency energy and a first low-frequency energy based on a first preset frequency range of the high-frequency band, a second preset frequency range of the low-frequency band, and the first frequency spectrum;

[0103] Calculating a first high-frequency to low-frequency voltage amplitude ratio based on the first high-frequency energy and the first low-frequency energy;

[0104] calculating a second high-frequency energy and a second low-frequency energy based on a first preset frequency range of the high-frequency band, a second preset frequency range of the low-frequency band, and the second frequency spectrum;

[0105] A second high-frequency and low-frequency voltage amplitude ratio is calculated based on the second high-frequency energy and the second low-frequency energy.

[0106] In this embodiment, key feature extraction also includes extracting frequency-domain energy ratio features. Frequency-domain energy ratio features are key to distinguishing fault types and assessing line status by quantifying the energy distribution characteristics of the signal in different frequency bands. Frequency-domain energy decomposition is performed on the traveling wave signal, and the ratio of high-frequency to low-frequency components (i.e., the high-frequency to low-frequency voltage amplitude ratio R) reflects the frequency-domain energy distribution characteristics of the transient traveling wave signal.

[0107] Specifically, FFT (Fast Fourier Transform) is performed on the first reconstructed signal and the second reconstructed signal to obtain spectrum X(f), which are recorded as the first spectrum and the second spectrum. Thereafter, the high-frequency energy and the low-frequency energy are calculated according to the first preset frequency range of the high-frequency band and the second preset frequency range of the low-frequency band, respectively:

[0108] The high-frequency energy calculation formula is:

[0109]

[0110] The formula for calculating low-frequency energy is:

[0111]

[0112] Then, the first high-frequency energy and the first low-frequency energy are amplitude-ratioed to obtain a first high-low frequency voltage amplitude ratio, and the second high-frequency energy and the second low-frequency energy are amplitude-ratioed to obtain a second high-low frequency voltage amplitude ratio, so as to quantify the frequency domain energy distribution characteristics of the transient signal.

[0113] The amplitude ratio expression is:

[0114]

[0115] Optionally, the first preset frequency range f of the high frequency band high =100kHz~1MHz; the second preset frequency range f of the low frequency band low=10kHz~100kHz.

[0116] In one embodiment of the present application, a specific key feature extraction scheme is provided. In S23, based on the first reconstructed signal, the second reconstructed signal, the first arrival times of the first modulus maximum points, and the second arrival times of the second modulus maximum points, feature values ​​of multiple target types are determined, which specifically includes:

[0117] In the first spectrum, determine the first frequency component with the largest energy, and record it as the first signal main frequency;

[0118] Calculate the theoretical wave velocity, and based on the main frequency of the first signal, perform dispersion effect correction on the theoretical wave velocity to obtain the first main frequency component wave velocity;

[0119] In the second spectrum, determine the second frequency component with the largest energy, and record it as the second signal main frequency;

[0120] Based on the main frequency of the second signal, the theoretical wave velocity is corrected for dispersion effect to obtain the second main frequency component wave velocity.

[0121] In this embodiment, key feature extraction also includes main frequency wave velocity feature extraction, which refers to accurately quantifying the propagation characteristics of traveling waves in distribution lines by analyzing the dominant frequency components of traveling wave signals and their corresponding propagation speeds. Perform FFT on the spectrum X(f) obtained and find the frequency component f corresponding to the maximum energy main =argmax f |X(f)| 2 , recorded as the first signal main frequency and the second signal main frequency.

[0122] Then, the theoretical wave speed is calculated based on the distribution line unit length parameters.

[0123] The theoretical wave velocity calculation formula is:

[0124]

[0125] Alternatively, assuming that the overhead line L = 0.8 mL / km, C = 12 nF / km, then v0 ≈ 2.98 × 10 5 km / s.

[0126] Then, consider the frequency-dependent resistance (Skin effect) Substitute f = f main , perform dispersion effect correction on v0 and obtain the corrected wave velocity v(f).

[0127] The dispersion effect correction expression is:

[0128]

[0129] For example, when f main When =300kHz, R0=0.1Ω / km, L=0.8mH / km, v(300kHz)≈0.97v0.

[0130] Through the above method, multi-dimensional feature expression is achieved, and various key feature quantities are comprehensively considered to reflect the severity of the fault from multiple angles, avoiding the limitations of single feature quantity evaluation. As a result, it can better adapt to this complex network environment and capture the subtle differences in transient traveling wave signals generated by faults at different locations, thereby achieving more accurate fault location.

[0131] S24: Mapping the eigenvalues ​​of various target types to the target interval using an interval normalization method to obtain characteristic parameters corresponding to the eigenvalues ​​of various target types.

[0132] S25: combining feature parameters of multiple target types in a preset order to generate a feature vector.

[0133] For steps S24-S25, interval normalization is used to uniformly map various eigenvalues ​​of different dimensions to specific intervals to obtain characteristic parameters corresponding to each type of eigenvalue. Subsequently, multiple characteristic parameters are combined to form a characteristic vector.

[0134] Optionally, the Min-Max normalization method is used to extract the global maximum value X from each type of eigenvalue max and minimum value X min , for each eigenvalue X of various types, calculate X according to the formula norm , to normalize each eigenvalue to the interval [-1, 1], and obtain the characteristic parameters corresponding to each type of eigenvalue:

[0135] The formula is:

[0136]

[0137] Afterwards, the five types of feature parameters are combined into a normalized feature vector F and stored in the matrix form F∈R N*5 (N is the number of samples), specifically, the preset order is: F = [Δt, A pu ,P,R,v], where Δt is the time difference of the traveling wave head, A pu is the transient surge amplitude per unit value, P is the transient surge polarity, R is the ratio of high-frequency and low-frequency voltage amplitudes, and V is the main frequency component wave velocity.

[0138] Through the above method, the convergence and accuracy of subsequent machine learning algorithms can be effectively improved.

[0139] S30: Based on the feature vector and preset condition information, a synthetic feature vector is generated through a conditional generative adversarial network.

[0140] In this step, in the diagnosis of distribution line faults, traditional methods usually rely on historical fault data for model training. However, the actual fault sample distribution is uneven, with too much common fault data and a scarcity of edge case samples (such as lightning strike-high resistance composite faults with an occurrence rate of <5%). This data imbalance problem causes the model to be severely biased towards high-frequency fault types during training, while the recognition ability of low-probability but high-risk complex faults (such as intermittent arcs, high-resistance grounding, etc.) is insufficient, affecting the accuracy and robustness of fault diagnosis. Since it is difficult to obtain enough edge fault samples in real scenarios, the embodiment of the present application proposes a distribution line fault feature parameter data enhancement method based on a generative adversarial network (GAN).

[0141] Specifically, typical fault samples, including short circuit faults, high-resistance ground faults, and lightning faults, were extracted from historical fault recording data to obtain feature vectors. A GAN network architecture was constructed, with pre-set conditional information (such as fault type, fault distance, and line parameters) used as generator constraints. The GAN was trained using the feature vectors of the fault samples to generate synthetic feature combinations consistent with the true distribution.

[0142] Optionally, GAN is trained using historical fault data (1000 groups) to generate 5000 sets of synthetic data so that the generator generates 5 feature combinations consistent with the true distribution.

[0143] In actual application scenarios, the GAN-based distribution line fault feature parameter data enhancement and synthesis method specifically includes: Conditional Generative Adversarial Network (CGAN) architecture design:

[0144] Generator Network (GENERATOR):

[0145] The network structure adopts a deep fully connected architecture:

[0146] Input layer: Receives a 100-dimensional random noise vector z~N(0,1) and preset condition information c (including fault distance x, fault type code, line parameters, etc.).

[0147] Hidden layer:

[0148] First fully connected layer: 256 neurons, ReLU activation function;

[0149] Conditional concatenation layer: concatenates the preset condition information c with the output of the first layer;

[0150] Second fully connected layer: 128 neurons, ReLU activation function;

[0151] Output layer: 5 neurons, Tanh activation function (corresponding to 5 feature parameters);

[0152] Output: Synthesized feature vector

[0153] Discriminator network (Discriminator):

[0154] Adopt conditional judgment architecture:

[0155] Input layer: receives the real feature vector X or the synthetic feature vector and preset condition information c.

[0156] Hidden layer:

[0157] First fully connected layer: 256 neurons, LeakyReLU activation (α=0.2);

[0158] Conditional concatenation layer: concatenates the preset condition information c with the output of the first layer;

[0159] Second fully connected layer: 64 neurons, LeakyReLU activation (α=0.2);

[0160] Output layer: 1 neuron, Sigmoid activation function;

[0161] Output: The probability of discriminating authenticity D(X,c)∈[0,1].

[0162] Optimization goal:

[0163] Loss function: L CGAN =E[logD(X,c)+E[log(1-D(G(z,c),c))]], where α~U(0,1) and λ=10 are the gradient penalty coefficients.

[0164] Adversarial training and synthetic data generation steps:

[0165] Training process:

[0166] Input: Real dataset

[0167] Batch training: Each batch randomly samples real data and noise to generate synthetic data.

[0168] Conditional injection: Inject preset condition information c into both the generator and the discriminator to ensure that the generated data is consistent with the conditions.

[0169] Training strategy: Gradient penalty: Using Wasserstein (WGAN-GP) to improve training stability:

[0170]

[0171] in Novel interpolation for real and synthetic data.

[0172] Learning rate scheduling: initial learning rate 10 -4 , decaying by 10% every 50 rounds.

[0173] Synthetic Data Generation:

[0174] Condition specification: Input target fault distance x target and line parameters to generate the corresponding feature vector.

[0175] Batch Generation: Generator Output After anti-normalization, the physical quantity is obtained.

[0176] Through this approach, we learn the distribution of real-world fault data and synthesize diverse fault scenario data, covering edge cases such as lightning-high-resistance combined faults that are difficult to capture using traditional methods, thereby addressing data scarcity. This enables multi-data fusion, improves feature representation, and expands the training set, enabling the model to maintain high recognition accuracy even in complex and rare fault scenarios. This addresses the computational challenges of traditional positioning methods, which rely on a single feature and struggle to cope with complex scenarios. Furthermore, GAN adversarial training can generate noisy synthetic data, improving the subsequent model's ability to resist interference and effectively distinguish between real signals and noise, enhancing its robustness against noise and complex scenarios.

[0177] S40: Constructing a long short-term memory model, and training the long short-term memory model based on the feature vector and the synthesized feature vector to obtain a distribution line fault location model.

[0178] In this step, the real feature vector and the synthetic feature vector are merged in proportion (such as 7:3) to construct a balanced training set, a long short-term memory model is constructed, and the model is trained through the training set to finally obtain a distribution line fault location model.

[0179] In actual application scenarios, the construction of a distribution line fault location model using a combination of time series signals and static features as input includes the following steps:

[0180] Timing signal channel:

[0181] Input data construction:

[0182] Timing characteristics: time series of the original transient signal (length T = 1000 points, sampling rate 1 MHz).

[0183] Static features: 5 feature parameters [Δt, A, P, R, v] are used as additional inputs.

[0184] Data format: input tensor∈R N*T*(1+5) (N = number of samples).

[0185] LSTM network structure:

[0186] Input layer: receives time series signals and static features, dimension = 6 (1 time series channel + 5 features).

[0187] LSTM layers: 2 layers, 64 units per layer, activation function = tanh, dropout rate = 0.3.

[0188] Fully connected layer: 32 units (ReLU) → output layer (linear activation, output fault distance x).

[0189] Loss function and optimization:

[0190] Composite loss function: mean square error (MSE) + feature consistency constraint:

[0191]

[0192] Here, λ = 0.1 balances the positioning error and feature matching.

[0193] Optimizer: Adam (learning rate = 0.001, decay rate = 0.95).

[0194] Through the above method, LSTM is used to capture the high-frequency characteristics of weak signals, thereby solving the problem of detecting rare faults (such as high-resistance grounding and intermittent arcs). LSTM is also used to model the temporal relationship of multiple reflected waves, and the LSTM-GAN online learning mechanism is used to dynamically update model parameters, reducing dependence on precise physical models. Compared with traditional positioning methods, it is less susceptible to noise interference and line parameter influences, improving the problem of high-resistance fault detection in distribution network lines.

[0195] In one embodiment of the present application, Figure 2 As shown in FIG, a multi-source data information distribution network overhead line fault location method based on LSTM-GAN is provided, which specifically includes the following steps:

[0196] Step 1: Collect transient traveling wave signals at the beginning and end of the distribution line and save the time series;

[0197] Step 2: Preprocess the transient traveling wave signal to obtain a reconstructed signal, wherein the preprocessing includes filtering, noise reduction, and unification.

[0198] Step 3: Extract the characteristic values ​​of various target types from the reconstructed signal, where the various target types include: traveling wave head time difference, transient surge amplitude characteristics, transient surge polarity characteristics, frequency domain energy ratio characteristics, and main frequency wave velocity characteristics;

[0199] Step 4: Normalize the eigenvalues ​​of various target types to obtain eigenvectors;

[0200] Step 5: Input the feature vector into the GAN network for data synthesis to obtain a synthetic feature vector;

[0201] Step 6: Based on the feature vector and the synthesized feature vector, train the distribution line fault location model based on the LSTM model, and input the real-time signal into the model to locate the fault location.

[0202] In this embodiment, the joint training based on LSTM-GAN specifically includes:

[0203] Two-stage training strategy:

[0204] Phase 1: Pre-training GAN: Use historical fault data (1,000 groups) to train GAN and generate 5,000 sets of synthetic data; the goal is to enable the generator to generate five feature combinations consistent with the true distribution.

[0205] Phase 2: Training the LSMT localization model: Input: real data (1000 groups) + synthetic data (5000 groups); Validation set: 20% of the data is used for early stopping.

[0206] Dynamic feature fusion:

[0207] Inject static features at each time step of LSTM: h t =LSTM(x t ,[Δt,A pu ,P,R,V]).

[0208] Positioning output and correction:

[0209] Basic positioning: LSTM directly outputs the fault distance x pred .

[0210] Dynamic correction of wave speed: according to the main frequency f main And the real-time environmental parameters update the wave velocity v and feed it back to the LSTM input.

[0211] Multi-end data fusion: If it is a dual-end measurement, the features of both ends are spliced ​​into a 10-dimensional input vector.

[0212] Anti-interference:

[0213] Noise injection training: Add Gaussian noise (SNR = 20dB) and pulse interference to the training data to improve the robustness of the model.

[0214] Through the above approach, a LSTM-GAN (Long Short-Term Memory Network-Generative Adversarial Network) model was jointly constructed. GAN was used to generate synthetic data, focusing on enhancing high-resistance fault data, effectively expanding the training sample and enabling the model to more fully learn and understand rare faults. Secondly, by leveraging the LSTM's ability to capture high-frequency features of weak signals, it can keenly detect subtle changes hidden in complex signals, effectively solving the detection challenges of rare faults such as high-resistance ground faults and intermittent arcs. Furthermore, LSTM was used to model the temporal relationships of multiple reflections. In distribution networks, traveling waves generated by faults are reflected at different locations, and the temporal relationships of these reflections contain rich information about the fault location. Modeling this with LSTM enables more accurate fault location analysis. Furthermore, LSTM-GAN features an online learning mechanism, allowing the model to dynamically update parameters based on real-time data. Traditional location methods typically rely on precise physical models, but in reality, line parameters can fluctuate due to various factors, resulting in increased location errors. However, the LSTM-GAN-based approach continuously adapts to new data and changing environments through online learning, providing a more reliable guarantee for the safe and stable operation of distribution networks.

[0215] S50: Acquire a real-time signal of the distribution line, and input the real-time signal into a distribution line fault location model to obtain a fault location of the distribution line.

[0216] In this step, sensors installed at both ends of the distribution line continuously collect real-time signals. These signals are then fed into the distribution line fault location model, which then performs calculations and inferences based on the input real-time signals, ultimately outputting the fault location information for the distribution line. The fault location information can be a specific distance (e.g., meters from the line's origin) or a specific section of the line.

[0217] As can be seen, in the above scheme, by integrating the multi-dimensional feature vectors of transient traveling wave signals, fault characteristics are fully explored. GANs are then used to simulate the distribution and characteristics of real data, enhancing scarce high-resistance fault data and expanding the training sample, enabling the model to learn a wider range of fault characteristic patterns. Subsequently, a long short-term memory (LSTM) network is used to train the distribution line fault location model. This effectively captures the high-frequency characteristics of weak signals generated by rare faults such as high-resistance grounding and intermittent arcing, solving the difficult problem of high-resistance fault detection in distribution network lines and significantly improving the accuracy of distribution line fault location.

[0218] In one embodiment, a distribution line fault location device is provided, which corresponds one-to-one to the distribution line fault location method in the above embodiment. Figure 3As shown, the distribution line fault location device 100 includes: a determination module 101, a first acquisition module 102, a feature extraction module 103, a first generation module 104, a second generation module 105, a second acquisition module 106 and a third generation module 107. The functional modules are described in detail as follows:

[0219] A determination module 101 is configured to determine a first transient traveling wave signal and a second transient traveling wave signal at the beginning and end of a distribution line based on historical fault recording data of the distribution line; a first acquisition module 102 is configured to remove power frequency interference and perform noise reduction processing on the first transient traveling wave signal and the second transient traveling wave signal to obtain a first reconstructed signal and a second reconstructed signal;

[0220] The feature extraction module 103 is used to extract feature values ​​from the first reconstructed signal and the second reconstructed signal, and normalize the extracted feature values ​​to obtain a feature vector.

[0221] A first generating module 104 is configured to generate a synthetic feature vector based on the feature vector and preset condition information by using a conditional generative adversarial network;

[0222] The second generating module 105 is used to construct a long short-term memory model, and train the long short-term memory model based on the feature vector and the synthetic feature vector to obtain a distribution line fault location model;

[0223] The second acquisition module 106 is used to acquire the real-time signal of the distribution line; the third generation module 107 is used to input the real-time signal into the distribution line fault location model to obtain the fault location of the distribution line.

[0224] In one embodiment, the feature extraction module 103 specifically includes:

[0225] Performing continuous wavelet transform on the first reconstructed signal and the second reconstructed signal respectively to generate a first time-frequency coefficient matrix and a second time-frequency coefficient matrix;

[0226] Based on the preset size, extracting multiple first modulus maximum points from the first time-frequency coefficient matrix and obtaining a first arrival time of each modulus maximum point, and extracting multiple second modulus maximum points from the second time-frequency coefficient matrix and obtaining a second arrival time of each modulus maximum point;

[0227] Determining characteristic values ​​of multiple target types based on the first reconstructed signal, the second reconstructed signal, multiple first arrival times of multiple first modulus maximum points, and multiple second arrival times of multiple second modulus maximum points, wherein the multiple target types include: traveling wave front time difference, transient surge amplitude characteristics, transient surge polarity characteristics, frequency domain energy ratio characteristics, and main frequency wave velocity characteristics;

[0228] The eigenvalues ​​of various target types are mapped to the target interval using the interval normalization method to obtain the characteristic parameters corresponding to the eigenvalues ​​of various target types;

[0229] The feature parameters of multiple target types are combined in a preset order to generate a feature vector.

[0230] In one embodiment, the feature extraction module 103 is further configured to:

[0231] Obtaining a first initial traveling wave arrival time of a first modulus maximum point among the plurality of first modulus maximum points, and a second initial traveling wave arrival time of a first modulus maximum point among the plurality of second modulus maximum points;

[0232] Calculating the absolute value of the difference between the first initial traveling wave arrival time and the second initial traveling wave arrival time to obtain the traveling wave front time difference;

[0233] Obtaining a first sign of a wavelet coefficient at a first initial traveling wave arrival time and a second sign of a wavelet coefficient at a second initial traveling wave arrival time;

[0234] determining a first transient surge polarity at an arrival time of a first initial traveling wave based on the first symbol;

[0235] A second transient surge polarity of the second initial traveling wave arrival time is determined based on the second sign.

[0236] In one embodiment, the feature extraction module 103 is further configured to:

[0237] Obtaining a first absolute amplitude corresponding to each first arrival time, and converting the first absolute amplitude into a first per-unit value to obtain a first transient surge amplitude;

[0238] A second absolute amplitude corresponding to each second arrival time is obtained, and the second absolute amplitude is converted into a second per-unit value to obtain a second transient surge amplitude.

[0239] In one embodiment, the feature extraction module 103 is further configured to:

[0240] Performing a fast Fourier transform on the first reconstructed signal to obtain a first spectrum;

[0241] Performing a fast Fourier transform on the second reconstructed signal to obtain a second spectrum;

[0242] Calculating a first high-frequency energy and a first low-frequency energy based on a first preset frequency range of the high-frequency band, a second preset frequency range of the low-frequency band, and the first frequency spectrum;

[0243] Calculating a first high-frequency to low-frequency voltage amplitude ratio based on the first high-frequency energy and the first low-frequency energy;

[0244] calculating a second high-frequency energy and a second low-frequency energy based on a first preset frequency range of the high-frequency band, a second preset frequency range of the low-frequency band, and the second frequency spectrum;

[0245] A second high-frequency and low-frequency voltage amplitude ratio is calculated based on the second high-frequency energy and the second low-frequency energy.

[0246] In one embodiment, the feature extraction module 103 is further configured to:

[0247] In the first spectrum, determine the first frequency component with the largest energy, and record it as the first signal main frequency;

[0248] Calculate the theoretical wave velocity, and based on the main frequency of the first signal, perform dispersion effect correction on the theoretical wave velocity to obtain the first main frequency component wave velocity;

[0249] In the second spectrum, determine the second frequency component with the largest energy, and record it as the second signal main frequency;

[0250] Based on the main frequency of the second signal, the theoretical wave velocity is corrected for dispersion effect to obtain the second main frequency component wave velocity.

[0251] In one embodiment, the second generating module 105 is specifically configured to:

[0252] Constructing long-short-term memory models;

[0253] The feature vector and the synthetic feature vector are input into the long short-term memory model, and the model parameters are back-propagated and gradient calculated based on the mean square error combined with the feature consistency constraint loss function. The model parameters are iteratively updated according to the gradient calculation results using the Adam optimizer;

[0254] When the change in the loss function is less than a preset threshold or the number of iterations is equal to the maximum number of iterations, the output model is used as the distribution line fault location model.

[0255] The present invention provides a distribution line fault location device 100. This device integrates the multidimensional feature vectors of transient traveling wave signals to fully explore fault characteristics. It also uses a Generative Adversarial Network (GAN) to simulate the distribution and characteristics of real data, enhance scarce high-resistance fault data, and expand the training sample, enabling the model to learn a wider range of fault characteristic patterns. Subsequently, a long short-term memory (LSTM) network is used to train the distribution line fault location model. This model effectively captures the high-frequency characteristics of weak signals generated by rare faults such as high-resistance grounding and intermittent arcing, solving the difficult problem of high-resistance fault detection in distribution network lines and significantly improving the accuracy of distribution line fault location.

[0256] The specific limitations of the distribution line fault locating device can be found in the limitations of the distribution line fault locating method described above and will not be further elaborated here. Each module in the above-described distribution line fault locating device can be implemented in whole or in part through software, hardware, or a combination thereof. Each of these modules can be embedded in or independent of a processor within an electronic device in hardware form, or stored in memory within the electronic device in software form, allowing the processor to call and execute the corresponding operations of each module.

[0257] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are performed:

[0258] Based on historical fault recording data of the distribution line, determining a first transient traveling wave signal and a second transient traveling wave signal at the beginning and end of the line, and performing power frequency interference removal and noise reduction processing on the first transient traveling wave signal and the second transient traveling wave signal to obtain a first reconstructed signal and a second reconstructed signal;

[0259] Extracting eigenvalues ​​of the first reconstructed signal and the second reconstructed signal, normalizing the extracted eigenvalues ​​and combining them to obtain a eigenvector;

[0260] Based on the feature vector and preset condition information, a synthetic feature vector is generated through a conditional generative adversarial network;

[0261] A long short-term memory model is constructed and trained based on the eigenvectors and synthetic eigenvectors to obtain a distribution line fault location model.

[0262] The real-time signal of the distribution line is obtained and input into the distribution line fault location model to obtain the fault location of the distribution line.

[0263] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0264] Based on historical fault recording data of the distribution line, determining a first transient traveling wave signal and a second transient traveling wave signal at the beginning and end of the line, and performing power frequency interference removal and noise reduction processing on the first transient traveling wave signal and the second transient traveling wave signal to obtain a first reconstructed signal and a second reconstructed signal;

[0265] Extracting eigenvalues ​​of the first reconstructed signal and the second reconstructed signal, normalizing the extracted eigenvalues ​​and combining them to obtain a eigenvector;

[0266] Based on the feature vector and preset condition information, a synthetic feature vector is generated through a conditional generative adversarial network;

[0267] A long short-term memory model is constructed and trained based on the eigenvectors and synthetic eigenvectors to obtain a distribution line fault location model.

[0268] The real-time signal of the distribution line is obtained and input into the distribution line fault location model to obtain the fault location of the distribution line.

[0269] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or electronic device can be referred to the relevant descriptions on the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0270] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0271] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0272] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A distribution network line fault location method, characterized in that: include: Based on historical fault recording data of the distribution network line, determine a first transient traveling wave signal and a second transient traveling wave signal at the beginning and end of the line, and perform power frequency interference removal and noise reduction processing on the first transient traveling wave signal and the second transient traveling wave signal to obtain a first reconstructed signal and a second reconstructed signal; Extracting eigenvalues ​​from the first reconstructed signal and the second reconstructed signal, normalizing the extracted eigenvalues, and combining them to obtain a eigenvector; Based on the feature vector and preset condition information, generating a synthetic feature vector through a conditional generative adversarial network; Constructing a long short-term memory model, and training the long short-term memory model based on the feature vector and the synthetic feature vector to obtain a distribution network line fault location model; A real-time signal of the distribution network line is obtained, and the real-time signal is input into the distribution network line fault location model to obtain the fault location of the distribution network line.

2. The method according to claim 1, characterized in that The step of extracting eigenvalues ​​from the first reconstructed signal and the second reconstructed signal, and normalizing the extracted eigenvalues ​​and combining them to obtain a eigenvector specifically includes: Performing continuous wavelet transform on the first reconstructed signal and the second reconstructed signal respectively to generate a first time-frequency coefficient matrix and a second time-frequency coefficient matrix; Based on a preset size, extracting a plurality of first modulus maximum points from the first time-frequency coefficient matrix and obtaining a first arrival time of each modulus maximum point, and extracting a plurality of second modulus maximum points from the second time-frequency coefficient matrix and obtaining a second arrival time of each modulus maximum point; Determining characteristic values ​​of multiple target types based on the first reconstructed signal, the second reconstructed signal, the multiple first arrival times of the multiple first modulus maxima points, and the multiple second arrival times of the multiple second modulus maxima points, wherein the multiple target types include: traveling wave front time difference, transient surge amplitude characteristic, transient surge polarity characteristic, frequency domain energy ratio characteristic, and main frequency wave velocity characteristic; The eigenvalues ​​of various target types are mapped to the target interval using the interval normalization method to obtain the characteristic parameters corresponding to the eigenvalues ​​of various target types; The feature parameters of multiple target types are combined in a preset order to generate a feature vector.

3. The method according to claim 2, characterized in that The step of determining characteristic values ​​of multiple target types based on the first reconstructed signal, the second reconstructed signal, the multiple first arrival times of the multiple first modulus maximum points, and the multiple second arrival times of the multiple second modulus maximum points specifically includes: Obtaining a first initial traveling wave arrival time of a first modulus maximum point among the plurality of first modulus maximum points, and a second initial traveling wave arrival time of a first modulus maximum point among the plurality of second modulus maximum points; Calculating the absolute value of the difference between the first initial traveling wave arrival time and the second initial traveling wave arrival time to obtain a traveling wave front time difference; Obtaining a first sign of a wavelet coefficient at the arrival time of the first initial traveling wave and a second sign of a wavelet coefficient at the arrival time of the second initial traveling wave; determining a first transient surge polarity at an arrival time of the first initial traveling wave based on the first symbol; A second transient surge polarity of the second initial traveling wave arrival time is determined based on the second sign.

4. The method according to claim 2, characterized in that The step of determining characteristic values ​​of multiple target types based on the first reconstructed signal, the second reconstructed signal, the multiple first arrival times of the multiple first modulus maximum points, and the multiple second arrival times of the multiple second modulus maximum points specifically further includes: Obtaining a first absolute amplitude corresponding to each first arrival time, and converting the first absolute amplitude into a first per-unit value to obtain a first transient surge amplitude; A second absolute amplitude corresponding to each second arrival time is obtained, and the second absolute amplitude is converted into a second per-unit value to obtain a second transient surge amplitude.

5. The method according to claim 2, characterized in that The step of determining characteristic values ​​of multiple target types based on the first reconstructed signal, the second reconstructed signal, the multiple first arrival times of the multiple first modulus maximum points, and the multiple second arrival times of the multiple second modulus maximum points specifically further includes: Performing a fast Fourier transform on the first reconstructed signal to obtain a first spectrum; Performing a fast Fourier transform on the second reconstructed signal to obtain a second spectrum; Calculating a first high-frequency energy and a first low-frequency energy based on a first preset frequency range of a high-frequency band, a second preset frequency range of a low-frequency band, and the first frequency spectrum; calculating a first high-frequency to low-frequency voltage amplitude ratio based on the first high-frequency energy and the first low-frequency energy; calculating a second high-frequency energy and a second low-frequency energy based on the first preset frequency range of the high-frequency band, the second preset frequency range of the low-frequency band, and the second frequency spectrum; A second high-frequency and low-frequency voltage amplitude ratio is calculated based on the second high-frequency energy and the second low-frequency energy.

6. The method according to claim 5, characterized in that The step of determining characteristic values ​​of multiple target types based on the first reconstructed signal, the second reconstructed signal, the multiple first arrival times of the multiple first modulus maximum points, and the multiple second arrival times of the multiple second modulus maximum points specifically further includes: In the first spectrum, determining a first frequency component with the largest energy, which is recorded as the first signal main frequency; Calculating a theoretical wave velocity, and performing dispersion effect correction on the theoretical wave velocity based on the main frequency of the first signal to obtain a first main frequency component wave velocity; In the second spectrum, determining a second frequency component with the maximum energy, and recording it as the second signal main frequency; Based on the main frequency of the second signal, the theoretical wave velocity is corrected for dispersion effects to obtain the second main frequency component wave velocity.

7. The method according to claim 1, characterized in that The step of constructing a long short-term memory model and training the long short-term memory model based on the feature vector and the synthetic feature vector to obtain a distribution network line fault location model specifically includes: Constructing the long short-term memory model; Inputting the feature vector and the synthesized feature vector into the long short-term memory model, performing backpropagation and gradient calculation on the model parameters based on the mean square error combined with the feature consistency constraint loss function, and iteratively updating the model parameters of the model according to the gradient calculation results using the Adam optimizer; When the change in the loss function is less than a preset threshold or the number of iterations is equal to the maximum number of iterations, the output model is used as the distribution network line fault location model.

8. A distribution network line fault location device, characterized in that: include: A determination module, configured to determine a first transient traveling wave signal and a second transient traveling wave signal at the beginning and end of a distribution network line based on historical fault recording data of the distribution network line; a first acquisition module, configured to perform power frequency interference removal and noise reduction processing on the first transient traveling wave signal and the second transient traveling wave signal to obtain a first reconstructed signal and a second reconstructed signal; a feature extraction module, configured to extract feature values ​​from the first reconstructed signal and the second reconstructed signal, and normalize the extracted feature values ​​to obtain a feature vector; A first generating module is configured to generate a synthetic feature vector based on the feature vector and preset condition information through a conditional generative adversarial network; A second generating module is configured to construct a long short-term memory model, and train the long short-term memory model based on the feature vector and the synthesized feature vector to obtain a distribution network line fault location model; A second acquisition module is used to obtain the real-time signal of the distribution network line; The third generating module is used to input the real-time signal into the distribution network line fault location model to obtain the fault location of the distribution network line.

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

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

Images