High-impedance fault detection method based on transient signal spectrum difference

Abstract

The invention discloses a high-impedance fault detection method based on transient signal spectrum differences, and belongs to the technical field of power distribution network fault detection. Transient zero-sequence current signals of all lines of a power distribution network are collected, and high-resolution spectrum characteristic spectrums corresponding to all the lines are generated through the separation step of signal subspaces and noise subspaces; establishing a frequency spectrum matrix difference quantification model, converting the frequency spectrum characteristic spectrum into a two-dimensional data matrix, and quantifying frequency spectrum difference characteristics among different lines by adopting a matrix norm calculation method to obtain a difference coefficient; based on a double-layer discrimination threshold system, setting a fault similarity coefficient as a first-layer discrimination threshold, and screening out a potential fault line combination; and setting a fault judgment coefficient as a second-layer judgment threshold value, performing difference comparison on the potential fault line and the normal line, and identifying a high-impedance fault line. The influence of normal transient disturbance such as load switching and capacitor bank switching is eliminated by using the characteristic of transient signal frequency spectrum difference.

Description

Technical Field

[0001] This invention relates to the field of power distribution network fault detection technology, and specifically to a high-impedance fault detection method based on transient signal spectral differences. Background Technology

[0002] High-impedance faults account for 5% to 10% of distribution network faults, with fault resistance reaching thousands of ohms, resulting in weak zero-sequence current amplitudes. Traditional protection devices based on steady-state zero-sequence current overcurrent are difficult to operate reliably, especially when high-impedance ground faults occur, as the fault characteristics are very weak and easily confused with the zero-sequence currents generated by normal transient disturbances such as capacitor switching and load switching.

[0003] Traditional overcurrent protection techniques cannot reliably detect high-resistance faults. In distribution networks, steady-state quantities are often used for fault detection, but these methods are limited by the neutral grounding method. As the transition resistance increases, the amplitude of the steady-state quantity decreases, making the extraction and identification of characteristic components difficult. Furthermore, the steady-state characteristic components detected during normal disturbances and faults can be confused.

[0004] While some scholars have proposed selecting fault feeders based on the polarity of transient quantities, this method is not only limited by the accuracy of current transformers, but also prone to false tripping due to normal transient disturbances in the distribution network. With the development of artificial intelligence algorithms, some scholars have proposed using artificial neural networks and support vector machines to identify the characteristic components of faults; however, the physical meaning of these methods is not clearly defined, and the accuracy of the detection results depends on the quality of the data.

[0005] High-resistance grounding faults can damage equipment insulation, disrupt the normal operation of the power grid, cause electric shock accidents, and lead to forest fires. Therefore, a fault detection method that can quickly and accurately identify faults and differentiate interferences is crucial for the normal and safe operation of the power distribution network. Summary of the Invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the technical problem solved by this invention is: how to solve the problem that traditional fault detection methods such as zero-sequence overcurrent protection fail in the case of high-impedance faults, eliminate the influence of the two normal transient disturbance conditions of capacitor bank switching and load switching, achieve accurate detection of high-impedance faults, improve the power grid's ability to identify high-impedance faults, and solve the technical problem that the characteristics of high-impedance faults are weak and easily confused with normal disturbances.

[0008] To solve the above technical problems, the present invention provides the following technical solution: a high impedance fault detection method based on transient signal spectrum differences, comprising: collecting transient zero-sequence current signals of each line of the distribution network; processing the transient zero-sequence current signals based on a signal subspace decomposition algorithm; and generating a high-resolution spectral feature map corresponding to each line through a separation step of signal subspace and noise subspace. A spectrum matrix-based difference quantification model is established, the spectrum feature map is converted into a two-dimensional data matrix, and the matrix norm calculation method is used to quantify the spectrum difference characteristics between different lines to obtain the difference coefficient. A fault similarity coefficient is set as the first-level discrimination threshold, and a fault determination coefficient is set as the second-level discrimination threshold to construct a two-level discrimination threshold system. Based on the two-level discrimination threshold system, potential faulty line combinations are screened by comparing the relationship between the difference coefficient and the fault similarity coefficient. For the selected combinations of potentially faulty lines, the spectral characteristics of the potentially faulty lines are compared with those of the normal lines. When the difference coefficient is greater than the fault determination coefficient, the line is identified as a high-impedance faulty line.

[0009] As a preferred embodiment of the high impedance fault detection method based on transient signal spectrum difference described in this invention, the method further includes: monitoring the zero-sequence voltage in the distribution network before acquiring the transient zero-sequence current signal of each line in the distribution network, and starting transient signal acquisition when the zero-sequence voltage exceeds the limit; the transient zero-sequence current signal includes transient signals generated by load switching, normal disturbance signals generated by capacitor bank switching, and fault signals generated by high impedance faults.

[0010] As a preferred embodiment of the high impedance fault detection method based on transient signal spectral differences described in this invention, the signal subspace decomposition algorithm includes obtaining the statistical characteristics of the signal through covariance matrix estimation. The signal space is divided into a signal subspace and a noise subspace based on eigenvalue decomposition; the orthogonality between the signal subspace and the noise subspace is used to generate a spectrum function with high frequency resolution.

[0011] As a preferred embodiment of the high impedance fault detection method based on transient signal spectrum difference described in this invention, the step of dividing the signal space into a signal subspace and a noise subspace based on eigenvalue decomposition includes: sorting the eigenvalues ​​by size, selecting eigenvalues ​​and eigenvectors equal to the number of signals D to form the signal subspace, and the remaining eigenvalues ​​and eigenvectors to form the noise subspace.

[0012] As a preferred embodiment of the high-impedance fault detection method based on transient signal spectral differences described in this invention, the high-resolution spectral feature map can distinguish the frequency domain distribution characteristics of normal disturbances and faults. The spectrum of normal disturbances exhibits local clustering distribution characteristics; the spectrum of high-impedance faults exhibits uniform distribution characteristics in the low-frequency and high-frequency bands, wherein the low-frequency band is in the normalized frequency range of 0 to 0.3 and the high-frequency band is in the normalized frequency range of 0.7 to 1.0.

[0013] As a preferred embodiment of the high-impedance fault detection method based on transient signal spectral differences described in this invention, the establishment of the spectral matrix difference quantification model includes: converting the three-dimensional spectral feature map into a two-dimensional grayscale data matrix; using the matrix norm as the difference measurement criterion to calculate the overall difference between the spectral matrices of different lines; and quantifying the comprehensive difference between spectral energy distribution and phase characteristics through the cumulative operation of the squared differences between matrix elements.

[0014] As a preferred embodiment of the high impedance fault detection method based on transient signal spectrum difference described in this invention, the dual-layer discrimination threshold system includes a first layer of discrimination based on fault similarity coefficient, used to distinguish between normal disturbance conditions and fault conditions, and to achieve preliminary screening of potential fault lines. The second layer of discrimination is based on the fault determination coefficient, which is used to identify high-impedance fault lines from potential fault lines, and to achieve the final fault location.

[0015] As a preferred embodiment of the high impedance fault detection method based on transient signal spectrum difference described in this invention, in the process of screening potential faulty line combinations, when the difference coefficient is less than the fault similarity coefficient, the corresponding line is determined to be a normal transient disturbance; the normal disturbance includes load switching and capacitor bank switching.

[0016] As a preferred embodiment of the high impedance fault detection method based on transient signal spectrum difference described in this invention, when the difference coefficient between two lines is greater than the fault similarity coefficient, it is determined that one of the two lines is a faulty line and the other line is a normal line, and the next step is executed; otherwise, both lines are determined to be normal transient disturbances, and the calculation process continues. The two are compared with the spectrum diagram of the normal line. When the difference coefficient is greater than the fault judgment coefficient, the line is judged to be a high impedance fault line.

[0017] As a preferred embodiment of the high impedance fault detection method based on transient signal spectral differences described in this invention, the signal subspace decomposition algorithm is a practical MUSIC algorithm, which constructs a noise matrix by calculating the estimated value of the covariance matrix and searching for spectral peaks by changing the incident angle of the signal source to obtain the spectral function.

[0018] The beneficial effects of this invention are as follows: The preliminary steps of this invention process transient zero-sequence current signals using a signal subspace decomposition algorithm to extract the frequency domain features of weak transient signals. The intermediate steps establish a spectrum matrix-based difference quantization model, which converts complex three-dimensional spectrum features into a quantifiable two-dimensional data matrix. Through calculation, it achieves accurate quantification of spectral differences between different lines, exhibiting better numerical stability and noise resistance compared to the traditional Euclidean distance method. The final step employs a dual-layer discrimination threshold system, using a hierarchical screening mechanism of fault similarity coefficient and fault judgment coefficient to solve the technical problem of weak high-impedance fault characteristics that are easily confused with normal disturbances. This enables accurate detection of high-impedance faults with transition resistances reaching thousands of ohms, eliminating the interference effects of normal transient disturbances such as load switching and capacitor bank switching. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 The above is a flowchart of a high-impedance fault detection method based on transient signal spectral differences, provided as an embodiment of the present invention.

[0021] Figure 2 The image shows the short-time Fourier spectrum of a high-impedance fault detection method based on transient signal spectral differences, provided as an embodiment of the present invention.

[0022] Figure 3 The image shows the short-time Fourier spectrum of capacitor bank switching in a high-impedance fault detection method based on transient signal spectral differences, provided as an embodiment of the present invention.

[0023] Figure 4 This is a short-time Fourier spectrum diagram of a high-impedance grounding fault in a high-impedance fault detection system based on transient signal spectral differences, provided as an embodiment of the present invention.

[0024] Figure 5 The image shows the load switching spectrum of a high impedance fault detection method based on transient signal spectrum differences, provided as an embodiment of the present invention.

[0025] Figure 6 The image shows the frequency spectrum of capacitor bank switching in a high impedance fault detection method based on transient signal frequency spectrum differences, as provided in an embodiment of the present invention.

[0026] Figure 7The spectrum diagram of a high-impedance grounding fault provided in an embodiment of the present invention is a high-impedance fault detection method based on transient signal spectrum difference.

[0027] Figure 8 This is a structural diagram of a 10kV distribution network model for a high-impedance fault detection method based on transient signal spectral differences, provided as an embodiment of the present invention.

[0028] Figure 9 This is a capacitor bank model diagram of a high-impedance fault detection method based on transient signal spectral differences, provided as an embodiment of the present invention.

[0029] Figure 10 This is a high-impedance grounding fault model diagram provided by an embodiment of the present invention for a high-impedance fault detection method based on transient signal spectrum difference. Detailed Implementation

[0030] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0031] Example 1, referring to Figures 1-4 This is one embodiment of the present invention, which provides a high-impedance fault detection method based on transient signal spectral differences, comprising: S100: Collects transient zero-sequence current signals from each line of the distribution network, processes the transient zero-sequence current signals based on the signal subspace decomposition algorithm, and generates high-resolution spectral feature maps corresponding to each line through the separation steps of signal subspace and noise subspace.

[0032] S200: Establish a spectrum matrix-based difference quantification model, convert the spectrum feature map into a two-dimensional data matrix, and use the matrix norm calculation method to quantify the spectrum difference characteristics between different lines to obtain the difference coefficient.

[0033] S300: Set the fault similarity coefficient as the first-level discrimination threshold and the fault determination coefficient as the second-level discrimination threshold to construct a two-level discrimination threshold system. Based on the two-level discrimination threshold system, potential faulty line combinations are screened by comparing the relationship between the difference coefficient and the fault similarity coefficient.

[0034] S400: For the selected combination of potential faulty lines, the spectral characteristics of the potential faulty lines are compared with those of the normal lines. When the difference coefficient is greater than the fault determination coefficient, the line is identified as a high-impedance faulty line.

[0035] It should be noted that the fault resistance of high-impedance faults in distribution networks can reach several thousand ohms, resulting in extremely weak zero-sequence current amplitudes, making them difficult for traditional steady-state zero-sequence current-based overcurrent protection devices to identify. In particular, when a high-impedance ground fault occurs, the resulting zero-sequence current signal is extremely similar in amplitude to the zero-sequence current generated by normal transient disturbances such as capacitor switching and load switching, easily causing protection devices to malfunction or fail to operate. Furthermore, the complex operating environment of distribution networks, with its various electromagnetic interferences and noise effects, further increases the technical difficulty of high-impedance fault detection, posing a serious challenge to the reliability and accuracy of traditional detection methods.

[0036] Therefore, to address the aforementioned challenges in high-impedance fault detection, a fault detection system based on transient signal spectral differences is constructed through steps S100-S400. This system utilizes a signal subspace decomposition algorithm to mine the deep frequency domain features of transient zero-sequence current signals, enabling accurate differentiation between normal disturbances and fault conditions. A spectral matrix-based difference quantification model is established to convert complex spectral features into quantifiable difference coefficients, providing numerical basis for fault identification. A dual-layer discrimination threshold system is designed, employing a hierarchical screening mechanism of fault similarity coefficients and fault determination coefficients to effectively eliminate the influence of normal transient disturbances, thereby achieving the identification of high-impedance faults and accurate location of faulty lines.

[0037] The technical terms used in this article are explained as follows: PSCAD (Power-Systems-Computer-Aided-Design) is a professional power system simulation software; MATLAB (Matrix-Laboratory) is a numerical computation and data analysis software platform; pmusic algorithm, or practical-MUSIC, is a practical version of the MUSIC algorithm; the MUSIC algorithm (Multiple-Signal-Classification) is a high-resolution spectrum estimation method that achieves accurate spectrum analysis through the orthogonality principle between the signal subspace and the noise subspace.

[0038] Example 2, refer to Figure 1 and Figure 10 As an embodiment of the present invention, a high-impedance fault detection method based on transient signal spectral differences is provided based on the previous embodiment, including: in step S100, transient zero-sequence current signals of each line in the distribution network are collected, the transient zero-sequence current signals are processed based on a signal subspace decomposition algorithm, and a high-resolution spectral feature map corresponding to each line is generated through the separation steps of signal subspace and noise subspace. Step S100 includes the following steps A1-A5: A1: Before collecting the transient zero-sequence current signals of each line in the distribution network, it also includes: monitoring the zero-sequence voltage in the distribution network, and starting transient signal acquisition when the zero-sequence voltage exceeds the limit; the transient zero-sequence current signal includes the transient signal generated by load switching, the normal disturbance signal generated by capacitor bank switching, and the fault signal generated by high impedance fault.

[0039] Specifically, the zero-sequence voltage in the distribution network is monitored first. When the zero-sequence voltage exceeds the limit, the threshold is usually set to 0.15Un. The transient zero-sequence current signal when the distribution network experiences disturbances or faults is collected in real time.

[0040] To illustrate the difference between faults and disturbances, this invention first uses the short-time Fourier transform function in MATLAB software to process the transient zero-sequence current signal to obtain a spectrum, thereby clarifying that there is a significant difference in the spectrum between faults and disturbances.

[0041] Figure 2 This is the short-time Fourier transform spectrum of the load switching. In power systems, the zero-sequence current generated by load switching, in addition to the fundamental frequency component, also contains harmonic components related to the power frequency due to the capacitive or inductive nature of the load. These harmonics are usually integer multiples of the fundamental frequency, such as the second harmonic of the fundamental frequency, and may also include other harmonic components, such as the fifth harmonic, seventh harmonic, etc. Using short-time Fourier transform to process the zero-sequence current waveform during load switching yields the following... Figure 2 The short-time Fourier spectrum is shown. Observing the bright blue bars in the graph reveals that during load switching, energy is concentrated at lower normalized frequencies, with very little energy distributed at higher frequencies, exhibiting an overall characteristic of a single, concentrated distribution.

[0042] Switching capacitor banks also generates transient zero-sequence current signals. Similar to load switching, the transient zero-sequence current signals generated by capacitor bank switching, in addition to the power frequency component, also have harmonic components. Second, third, or fifth harmonic components may appear.

[0043] Such frequency characteristics will cause the capacitor bank's spectrum to exhibit features similar to those of load switching, with energy concentration at several specific normalized frequencies, particularly at lower normalized frequencies. This concentration will appear discretely on the overall spectrum. This characteristic is readily apparent in the spectrum obtained by processing the zero-sequence current generated during capacitor bank switching using a short-time Fourier transform. Furthermore, a sharp increase and decrease in energy can be observed at certain normalized frequencies. Figure 3 This is reflected in the short-time Fourier spectrum: two bright yellow bars appear at the normalized frequencies, indicating a higher content at these locations.

[0044] Unlike load switching and capacitor bank switching, the transient current signal spectrum during a high-resistance ground fault exhibits a relatively uniform distribution across both high and low frequency bands. Compared to load switching and capacitor bank switching, its spectrum shows a denser and richer distribution in the high-frequency range of the normalized frequency. This spectral characteristic can be observed in the short-time Fourier spectrum of a high-resistance ground fault, such as... Figure 4 As shown in the figure, from lower to higher normalized frequencies, a distribution of larger values, represented by light green bands, appears. Close observation reveals several clusters of darker bars within these bands.

[0045] Based on the above analysis of the frequency characteristics under disturbance conditions and high-resistance grounding faults, the spectrum diagrams obtained after processing the transient zero-sequence current signals during load switching, capacitor bank switching, and high-resistance grounding faults using a spectrum analysis algorithm show differences in distribution and energy amplitude. Therefore, the differences in the transient signal spectrum can be analyzed based on the idea of ​​a difference identification algorithm, thereby distinguishing between normal disturbance conditions and high-resistance grounding faults, ultimately achieving the detection of high-resistance grounding faults.

[0046] The spectrum obtained by short-time Fourier transform only describes the energy distribution and is insufficient to express the difference between interference and faults. The pmusic algorithm, a high-resolution spectrum estimation method, can clearly present the content of different frequency bands of a signal and can even describe the trend changes in energy distribution. This helps to highlight the differences between spectrum diagrams, and therefore mathematical methods can be used to describe the differences between spectrum diagram features.

[0047] A2: Signal subspace decomposition algorithms include obtaining the statistical properties of the signal through covariance matrix estimation; The signal space is divided into a signal subspace and a noise subspace based on eigenvalue decomposition; the orthogonality between the signal subspace and the noise subspace is used to generate a spectrum function with high frequency resolution.

[0048] Specifically, the pmusic algorithm, characterized by its high resolution, is used for spectral analysis of the signal. This algorithm overcomes the spectral leakage problem of traditional algorithms by constructing signal and noise subspaces. The generated spectrum clearly reveals the frequency domain characteristics under different operating conditions: the spectrum of normal transient disturbances exhibits obvious local clustering characteristics, characterized by concentrated distribution in the low-frequency or high-frequency bands. In contrast, the spectrum of high-impedance faults shows a more uniform distribution in both the low-frequency and high-frequency bands.

[0049] A3: The signal subspace decomposition algorithm is a practical MUSIC algorithm. It constructs a noise matrix by calculating the estimated value of the covariance matrix and searches for spectral peaks by changing the incident angle of the signal source to obtain the spectral function. The method is applicable to distribution network systems with ungrounded neutral points or grounded through arc suppression coils.

[0050] Specifically, the basic programming principle of the pmusic algorithm is briefly described as follows: For the obtained N signal vectors, estimate the covariance matrix as shown in formula (1).

[0051] (1) in, is the covariance matrix; N is the number of signal vectors; Let i be the vector of the i-th snapshot; The output matrix is ​​the conjugate transpose. Its eigenvalues ​​are then decomposed, and its subspace is partitioned, as shown in equation (2).

[0052] (2) Where V is the eigenvector matrix; This is the conjugate transpose of the eigenvector matrix; These are the eigenvalues ​​of the first eigenvector; Let be the eigenvalue of the Lth eigenvector; diag is the diagonal element extracted from the matrix; It is a diagonal eigenvalue matrix.

[0053] The first D largest eigenvalues ​​correspond to the real signal components, and the remaining MD smaller eigenvalues ​​correspond to the noise components. The noise subspace matrix En is composed of the last MD eigenvectors, i.e., formula (5). This matrix is ​​orthogonal to the steering vector of the real signal and is the theoretical basis for the high-resolution characteristics of the MUSIC algorithm. In the peak search stage, the spectral function formula (6) is calculated by changing the incident angle θ of the signal source, where a(θ) is the steering vector. When θ is close to the direction of the real signal, the denominator approaches zero, and the spectral function shows a sharp peak, thus achieving high-precision frequency estimation. Compared with the traditional FFT method, this algorithm can achieve super-resolution spectral analysis under short data length conditions, and is particularly suitable for the accurate frequency domain feature extraction of transient signals.

[0054] A4: Dividing the signal space into a signal subspace and a noise subspace based on eigenvalue decomposition includes sorting the eigenvalues ​​by size, selecting eigenvalues ​​and eigenvectors equal to the number of signals D to form the signal subspace, and the remaining eigenvalues ​​and eigenvectors to form the noise subspace.

[0055] Specifically, based on the order of the eigenvalues, the largest eigenvalue and eigenvector, which are equal to the number of signals D, are divided into signal spaces E. s The remaining eigenvalues ​​and eigenvectors are then divided into a noise space E.n .

[0056] The partitioning process is shown in formula (3), where the signal matrix Es and the noise matrix E n As shown in formulas (4) and (5).

[0057] (3) (4) (5) Among them, A H v is the conjugate transpose of the matrix. i Here, D is the number of signal sources, and M represents the order of the matrix. For signal matrix; This is the first eigenvector of the signal matrix after sorting by eigenvalues. This is the D-th eigenvector of the signal matrix; This is the noise matrix; This is the first eigenvector in the noise matrix after sorting by eigenvalue. This is the second eigenvector in the noise matrix; This is the third eigenvector in the noise matrix; Let Q be the Qth eigenvector.

[0058] Finally, by changing the incident angle of the signal source and finding the peak value of the spectral function, the result is obtained as shown in formula (6).

[0059] (6) Among them, P mu (θ) is the obtained spectral function, a(θ) is the steering vector, and E n This is the noise matrix; This is the conjugate transpose of the guiding vector; This is the conjugate transpose of the noise matrix.

[0060] A5: High-resolution spectral feature maps can distinguish the frequency domain distribution characteristics of normal disturbances and faults. The spectrum of normal disturbances shows local clustering distribution characteristics; the spectrum of high-impedance faults shows uniform distribution characteristics in the low-frequency and high-frequency bands, with the low-frequency band ranging from 0 to 0.3 normalized frequencies and the high-frequency band ranging from 0.7 to 1.0 normalized frequencies.

[0061] Specifically, the pmusic algorithm in MATLAB was used to perform spectral analysis of the signal. The spectrum generated by this algorithm clearly reveals the frequency domain characteristics under different operating conditions: the spectrum of normal transient disturbances shows obvious local clustering characteristics, characterized by concentrated distribution in the low-frequency or high-frequency range. In contrast, the spectrum of high-impedance faults shows a more uniform distribution in both the low-frequency and high-frequency ranges. Figures 5-7 As shown.

[0062] In this embodiment, step S200 establishes a spectrum matrix-based difference quantization model, including the following steps B1-B3: B1: Convert the three-dimensional spectral feature map into a two-dimensional grayscale data matrix; Specifically, firstly, the 3D spectrogram generated by the pmusic algorithm contains three dimensions: frequency, time, and amplitude. The frequency axis reflects the signal's frequency domain distribution characteristics, the time axis reflects the evolution of transient processes, and the amplitude axis characterizes the energy intensity of each frequency component. During the conversion process, a projection mapping method is used to compress the 3D information into a 2D plane. By setting the viewing angle and projection parameters, key frequency domain features are ensured to be preserved. Gray-scale mapping employs a linear normalization strategy to reduce the spectral amplitude range... Mapping to the grayscale range [0, 255], the mapping formula is: in, The original amplitude. For the corresponding grayscale value; This represents the minimum value of the spectral amplitude. This represents the maximum amplitude of the spectrum. To enhance the distinguishability of spectral features under different operating conditions, a contrast enhancement technique is also employed. By adjusting the nonlinear parameters of the grayscale mapping, detailed features in the medium amplitude range are highlighted. The transformed two-dimensional matrix maintains the spatial topology of the original spectrum diagram. The row indices of the matrix correspond to the frequency axis, the column indices correspond to the time axis, and the matrix element values ​​correspond to the energy intensity at the corresponding time frequency point.

[0063] B2: Using matrix norm as a difference measure, calculate the overall difference between the spectrum matrices of different lines; B3: The combined difference between spectral energy distribution and phase characteristics is quantified by accumulating the squared differences between matrix elements.

[0064] Specifically, the three-dimensional spectrogram is converted into a two-dimensional grayscale matrix A in MATLAB software by calling a file. The spectrogram p1 corresponding to line 1 is taken, and the F-norm difference value between the matrices converted from spectrogram p1 and spectrogram pi is calculated. The difference between the matrices is calculated using the Frobenius norm and defined as the difference coefficient D. in, This is the grayscale matrix converted from the first spectrogram; This is the grayscale matrix transformed from the second spectrogram; F is the standard notation of the F-norm; The element in the i-th row and j-th column of the first matrix; This refers to the element in the i-th row and j-th column of the second matrix; The Frobenius normal form and the calculation of the difference coefficient D are shown in formulas (7) and (8): (7) (8) in, Let A be the transpose of matrix A; m and n are the number of rows and columns of the matrix, respectively. Let A be the trace of matrix A, which is the sum of the elements on the main diagonal of the matrix.

[0065] In this embodiment, step S300, based on a two-layer discrimination threshold system, includes the following steps C1-C2: C1: The first layer of discrimination is based on the fault similarity coefficient, which is used to distinguish between normal disturbance conditions and fault conditions, and to achieve the initial screening of potential faulty lines; The second layer of discrimination is based on the fault determination coefficient, which is used to identify high-impedance fault lines from potential fault lines, and to achieve the final fault location.

[0066] C2: When the difference coefficient is less than the fault similarity coefficient, the corresponding line is judged as a normal transient disturbance; the fault similarity coefficient is used to distinguish between normal disturbances and faults; normal disturbances include load switching and capacitor bank switching.

[0067] Specifically, the design of the two-layer discrimination threshold system in C1 is based on statistical analysis and theoretical derivation of a large amount of experimental data. The determination of the fault similarity coefficient FS first requires collecting difference coefficient samples for different types of normal disturbance conditions, including data from various load switching, capacitor bank switching, and other typical disturbance scenarios. Through statistical analysis of these samples, the sample mean μ and standard deviation σ are calculated. Assuming the difference coefficients follow a normal distribution, then: Where FS is the fault similarity coefficient; k is the confidence coefficient; μ is the sample mean; and σ is the standard deviation. Through ROC curve analysis and cross-validation, it was determined that k=2.5 maximizes the sensitivity of fault detection while ensuring 99.7% accuracy in detecting normal disturbances. The physical significance of this threshold lies in establishing a boundary between normal disturbances and fault conditions in terms of spectral differences, utilizing the essential difference between the local clustering characteristics of the normal disturbance spectrum and the uniform distribution characteristics of the fault spectrum. The second-layer fault judgment coefficient FD is set based on the distribution law of spectral differences between faulty and normal lines. Through extensive simulation experiments under different transition resistance conditions, the distribution range of the difference coefficient under fault conditions was statistically analyzed, and a lower bound that can cover more than 95% of fault samples was selected as the FD value. The synergistic mechanism of the two-layer threshold achieves an organic combination of coarse screening and fine discrimination. The first layer quickly eliminates obvious normal disturbances, reducing computational load and false alarm rate, while the second layer identifies high-impedance faults.

[0068] In this embodiment, step S400 sets a fault determination coefficient as a second-layer discrimination threshold for the selected potential faulty line combinations, including the following step D1: D1: When the difference coefficient between two lines is greater than the fault similarity coefficient, it is determined that one of the two lines is a faulty line and the other line is a normal line, and the next step is executed; otherwise, both lines are determined to be normal transient disturbances and the calculation process continues. The two are compared with the spectrum diagram of the normal line. When the difference coefficient is greater than the fault judgment coefficient, the line is judged to be a high impedance fault line.

[0069] Specifically, the complete process of fault line identification in D1 includes multiple verification steps and technical details. After potential faulty line combinations are screened out through the first layer of discrimination, a cross-validation strategy is needed for precise location. During the verification process, the spectrum of each potential faulty line is compared with the spectrum of all confirmed normal lines in the system using F-normative difference calculations to obtain a set of difference coefficient vectors: in, The coefficient of variation; The F-normative difference coefficients between the first normal line and the line to be judged are the difference coefficient vectors. The first element in; The F-normative difference coefficient between the second normal line and the line to be judged is the difference coefficient vector. The second element in the equation; k represents the number of normal lines. To eliminate the influence of special operating conditions of individual lines on the judgment results, a statistical averaging method is used to calculate the comprehensive difference coefficient: in, The coefficient of variation is denoted by g; g represents the number of normal lines. Let F-norm difference coefficient be the difference coefficient between the i-th normal line and the line to be judged, and let F be the difference coefficient vector. The i-th element in the array, where i is the index number of the normal line.

[0070] This value can more comprehensively reflect the degree of difference between the fault-prone line and the entire group of normal lines. The determination of the fault judgment coefficient FD=8.67 is based on a large number of experimental verifications. By setting high-impedance faults with different locations and resistance values ​​in a 10kV distribution network simulation model, the distribution characteristics of the difference coefficient between the faulty line and the normal line were statistically analyzed.

[0071] This quantization method preserves the energy distribution information of the spectrum and reflects the phase characteristics of the frequency components, exhibiting better stability than the traditional Euclidean distance metric.

[0072] Take the spectrum corresponding to line 1 Figure 2 Calculate the F-norm difference between the matrices transformed from spectrum p1 and spectrum pi, and define this value as the difference coefficient D. If the difference coefficient D between the two lines is greater than the fault similarity coefficient FS, then one of them is determined to be a faulty line and the other to be a normal line. Next, the spectrum diagrams of both lines are compared with those of the normal line. If the difference coefficient D is greater than the fault determination coefficient FD, then the line is determined to be a high-impedance faulty line. Otherwise, it is a normal line. The fault detection process ends, and fault line selection is achieved. Otherwise, both lines are determined to be normal transient disturbances, and the calculation process continues.

[0073] After analyzing multiple sets of experimental data, the fault detection initiation criterion FS was set to 6 to distinguish between normal disturbances and faults; the fault determination coefficient FD was set to 8.67 to identify high-impedance fault lines. Experiments showed that when the transition resistance was in the range of 1-10kΩ, the average difference coefficient between faulty and normal lines ranged from 8.95 to 9.91, while the difference coefficient under normal disturbance conditions was typically less than 6. Choosing 8.67 as the determination threshold ensured reliable detection of 10kΩ high-impedance faults while avoiding misjudgments of normal disturbances. This threshold also considered detection stability under different signal-to-noise ratio conditions, maintaining good discrimination performance even under adverse conditions where the signal-to-noise ratio dropped to 40dB.

[0074] In summary, during the signal processing stage, the practical MUSIC algorithm, compared to the traditional short-time Fourier transform method, enables super-resolution spectral analysis under short data length conditions, making it particularly suitable for accurate frequency domain feature extraction of transient processes in distribution networks. By separating the signal subspace from the noise subspace, the true signal components are separated from noise interference, enhancing the anti-interference capability of signal processing. In the spectral feature quantization stage, the conversion from a 3D spectrogram to a 2D grayscale matrix preserves complete frequency domain spatial structure information. Linear normalization and contrast enhancement techniques ensure the comparability and discriminability of spectral features under different operating conditions. The matrix processing method transforms the complex spectral analysis problem into a standard numerical calculation problem, facilitating engineering implementation and computational optimization. In the fault identification stage, the design of the dual-layer threshold system is based on extensive statistical data and theoretical analysis. The determination of the fault similarity coefficient uses statistical methods to ensure that the misjudgment rate of normal disturbances is controlled within a reasonable range. The setting of the fault judgment coefficient is based on experimental verification, enabling the identification of high-impedance faults with different resistance ranges. The cross-validation mechanism eliminates the influence of special operating conditions on individual lines through multiple comparisons and statistical averaging.

[0075] Example 3, referring to Figure 1 and Figure 10 This invention provides a high-impedance fault detection method based on transient signal spectral differences. To verify the beneficial effects of this invention, scientific demonstration is conducted through experiments.

[0076] To verify the feasibility of this invention, a model was constructed using PSCAD. Figure 8 The 10kV distribution network model shown has F1, F2, F3, and F4 as the set points for high-resistance grounding faults. In the simulation, the transition resistance of the high-resistance grounding fault can be set to 1-10kΩ. The capacitor bank and high-resistance fault model are as follows... Figure 9 and Figure 10 As shown in Table 1, the relevant parameters of the power distribution network are as follows: Table 1 Distribution Network Parameters

[0077] The transient zero-sequence currents under load switching, capacitor bank switching, and a 10kΩ high-resistance ground fault at point F1 were obtained using PSCAD simulation software, and the data were imported into MATLAB software. Differences were calculated between the transient disturbance conditions and the transient zero-sequence current spectrum under the high-resistance ground fault conditions. The results of the difference analysis between the spectrum under the high-resistance ground fault and the spectra under load switching and capacitor bank switching are shown in Table 2. Table 2 Detection results of transient disturbance conditions

[0078] exist Figure 7After setting a high-impedance fault at F1 on line L1, the zero-sequence current signals collected at the beginning of each line were processed to obtain a spectrum diagram. Then, the spectrum diagrams of the faulty lines were compared with those of the healthy lines to identify differences. The results are shown in Table 3. Table 3. Difference coefficients between faulty lines and normal lines

[0079] High-resistance faults with different transition resistances were set at four fault setting points in the distribution network model. The feasibility of fault detection was verified by simulation, and the results are shown in Table 4. Table 4. Detection results of faults at different locations

[0080] Simulation results show that the ground fault detection method based on the difference in transient signal spectrum can accurately detect high-resistance ground faults with a transition resistance of up to 10kΩ. Multiple different fault points can effectively distinguish between faulty lines and healthy lines, verifying the adaptability of the method.

[0081] During fault detection, the signals collected in the actual power distribution network contain a certain amount of noise. Therefore, noise is added as interference in the simulation to analyze the impact of different signal-to-noise ratios on fault detection. The simulation results are shown in Table 5. Table 5 Fault detection results under different signal-to-noise ratios

[0082] Although noise affects the content of characteristic frequency components in the spectrum, the spectral difference between the faulty line and the normal line is still significant. Simulations verify that the fault detection method proposed in this invention has a certain anti-interference capability.

[0083] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A high-impedance fault detection method based on transient signal spectral differences, characterized in that: This includes collecting transient zero-sequence current signals from each line of the distribution network, processing the transient zero-sequence current signals based on a signal subspace decomposition algorithm, and generating high-resolution spectral feature maps corresponding to each line through a separation step between the signal subspace and the noise subspace. A spectrum matrix-based difference quantification model is established, the spectrum feature map is converted into a two-dimensional data matrix, and the matrix norm calculation method is used to quantify the spectrum difference characteristics between different lines to obtain the difference coefficient. A fault similarity coefficient is set as the first-level discrimination threshold, and a fault determination coefficient is set as the second-level discrimination threshold to construct a two-level discrimination threshold system. Based on the two-level discrimination threshold system, potential faulty line combinations are screened by comparing the relationship between the difference coefficient and the fault similarity coefficient. For the selected combinations of potentially faulty lines, the spectral characteristics of the potentially faulty lines are compared with those of the normal lines. When the difference coefficient is greater than the fault determination coefficient, the line is identified as a high-impedance faulty line.

2. The high-impedance fault detection method based on transient signal spectral differences as described in claim 1, characterized in that: Before collecting transient zero-sequence current signals from each line of the distribution network, the zero-sequence voltage in the distribution network is monitored. When the zero-sequence voltage exceeds the limit, transient signal acquisition is initiated. The transient zero-sequence current signals include transient signals generated by load switching, normal disturbance signals generated by capacitor bank switching, and fault signals generated by high-impedance faults.

3. The high-impedance fault detection method based on transient signal spectral differences as described in claim 2, characterized in that: The signal subspace decomposition algorithm includes obtaining the statistical characteristics of the signal through covariance matrix estimation; Based on eigenvalue decomposition, the signal space is divided into a signal subspace and a noise subspace; By utilizing the orthogonality principle between the signal subspace and the noise subspace, a spectrum function with high frequency resolution is generated.

4. The high-impedance fault detection method based on transient signal spectral differences as described in claim 3, characterized in that: The step of dividing the signal space into a signal subspace and a noise subspace based on eigenvalue decomposition includes: sorting the eigenvalues ​​from smallest to largest; selecting eigenvalues ​​and eigenvectors equal to the number of signals D to form the signal subspace; and using the remaining eigenvalues ​​and eigenvectors to form the noise subspace.

5. The high-impedance fault detection method based on transient signal spectral differences as described in claim 4, characterized in that: The high-resolution spectral feature map can distinguish the frequency domain distribution characteristics of normal disturbances and faults. The spectrum of normal disturbances shows local clustering distribution characteristics; the spectrum of high impedance faults shows uniform distribution characteristics in the low-frequency and high-frequency bands, where the low-frequency band is the normalized frequency range of 0 to 0.3 and the high-frequency band is the normalized frequency range of 0.7 to 1.

0.

6. The high-impedance fault detection method based on transient signal spectral differences as described in claim 5, characterized in that: The establishment of the spectrum matrix differential quantization model includes converting the three-dimensional spectrum feature map into a two-dimensional grayscale data matrix; The matrix norm is used as the difference criterion to calculate the overall difference between the spectrum matrices of different lines; the comprehensive difference between the spectrum energy distribution and phase characteristics is quantified by the cumulative operation of the squared differences between matrix elements.

7. The high-impedance fault detection method based on transient signal spectral differences as described in claim 6, characterized in that: The dual-layer discrimination threshold system includes a first layer of discrimination based on the fault similarity coefficient, which is used to distinguish between normal disturbance conditions and fault conditions, and to achieve preliminary screening of potential fault lines. The second layer of discrimination is based on the fault determination coefficient, which is used to identify high-impedance fault lines from potential fault lines, and to achieve the final fault location.

8. The high-impedance fault detection method based on transient signal spectral differences as described in claim 7, characterized in that: In the process of screening potential faulty line combinations, when the difference coefficient is less than the fault similarity coefficient, the corresponding line is judged as a normal transient disturbance; the normal disturbance includes load switching and capacitor bank switching.

9. The high-impedance fault detection method based on transient signal spectral differences as described in claim 8, characterized in that: When the difference coefficient between two lines is greater than the fault similarity coefficient, it is determined that one of the two lines is a faulty line and the other line is a normal line, and the next step is executed; otherwise, both lines are determined to be normal transient disturbances and the calculation process continues. The two are compared with the spectrum diagram of the normal line. When the difference coefficient is greater than the fault judgment coefficient, the line is judged to be a high impedance fault line.

10. The high-impedance fault detection method based on transient signal spectral differences as described in claim 9, characterized in that: The signal subspace decomposition algorithm is a practical MUSIC algorithm. It constructs a noise matrix by calculating the estimated value of the covariance matrix and searches for spectral peaks by changing the incident angle of the signal source to obtain the spectral function.

Images