High resistance ground fault detection method and device based on fault equivalent admittance trajectory, equipment and medium

Abstract

The application discloses a high-resistance grounding fault detection method and device based on fault equivalent admittance trajectory, equipment and medium, and relates to the technical field of electrical measurement. The method comprises the following steps: collecting zero sequence voltage and zero sequence current at the head of each feeder, dividing a window and constructing an analytical signal; estimating background zero sequence admittance and background current according to a background window, and obtaining a fault residual current after deduction; constructing a fault equivalent admittance trajectory according to the residual current and the zero sequence voltage, and obtaining a fault starting mode weight through time sequence segmentation; extracting an arc growth slope in multiple scales, combining a resonance damping coefficient to obtain a damping normalized arc growth slope; fitting an arc growth model and a normal switching damping model to obtain a model competition discriminant; extracting a frequency band energy dispersion degree and a mode sequence consistency variable, combining the foregoing slope and the model competition discriminant to generate a detection result. According to the application, the bus zero sequence voltage and the zero sequence current at the head of each feeder can be used to distinguish high-resistance grounding faults from normal switching transients.

Description

Technical Field

[0001] This application relates to the field of electrical measurement technology, and in particular to a method, apparatus, equipment and medium for detecting high-resistance grounding faults based on fault equivalent admittance trajectories. Background Technology

[0002] When a high-resistance grounding fault occurs in a distribution network, the fault current is usually small and easily affected by changes in the grounding medium, the arc development process, and system resonance damping. To promptly identify the faulty feeder, it is necessary to use the bus zero-sequence voltage and the zero-sequence current at the beginning of each feeder to identify the high-resistance grounding fault.

[0003] Currently, the amplitude, phase, transient energy, or harmonic characteristics of zero-sequence voltage and zero-sequence current are usually analyzed, and fault detection results are generated based on preset thresholds or characteristic differences between different feeders.

[0004] However, normal switching operations also cause transient changes in zero-sequence voltage and zero-sequence current, whose signal characteristics may be similar to those of high-resistance grounding faults. Simultaneously, the feeder background zero-sequence admittance and system resonant damping affect the detection characteristics, making it difficult for existing methods to reliably distinguish between high-resistance grounding faults and normal switching transients. Therefore, how to distinguish between high-resistance grounding faults and normal switching transients based on the bus zero-sequence voltage and the zero-sequence current at the beginning of each feeder has become an urgent problem to be solved. Summary of the Invention

[0005] The purpose of this application is to provide a method, device, equipment and medium for detecting high-resistivity grounding faults based on fault equivalent admittance trajectories, aiming to solve the technical problem of how to distinguish between high-resistivity grounding faults and normal switching transients based on bus zero-sequence voltage and zero-sequence current at the beginning of each feeder.

[0006] To achieve the above objectives, this application proposes a high-resistivity grounding fault detection method based on fault equivalent admittance trajectories, the method comprising: Obtain the zero-sequence voltage and the zero-sequence current at the beginning of each feeder, divide the detection window, and construct the zero-sequence voltage analytical signal and the zero-sequence current analytical signal at the beginning of each feeder; Based on the background window, estimate the background zero-sequence admittance and background current analytical signal of each feeder, and subtract the background current analytical signal corresponding to each feeder from the zero-sequence current analytical signal at the beginning of each feeder to obtain the fault residual current analytical signal. Based on the fault residual current analytical signal and the zero-sequence voltage analytical signal, the fault equivalent admittance trajectory of each feeder is constructed, and the fault equivalent admittance trajectory is input into the time-series segmentation model to obtain the fault initiation morphology weight. At each preset time scale, an enhanced conductance logarithmic value is generated from the real part of the fault equivalent admittance trajectory. The arc growth slope is extracted according to the fault initiation morphology weight, the resonance damping coefficient is obtained, and the damped normalized arc growth slope is obtained according to the arc growth slope and the resonance damping coefficient. The model competition discriminant is obtained based on the fitting results of the arc growth model and the normal shear damping model; Based on the fault equivalent admittance trajectory, frequency band energy dispersion and morphological order consistency variables are extracted, and detection results are generated by combining the damped normalized arc growth slope and the model competition discriminant.

[0007] Furthermore, to achieve the above objectives, this application also proposes a high-resistance grounding fault detection device based on fault equivalent admittance trajectory, the device comprising: The data processing module is used to acquire zero-sequence voltage and zero-sequence current at the beginning of each feeder, divide the detection window, and construct the zero-sequence voltage analytical signal and the zero-sequence current analytical signal at the beginning of each feeder. The residual generation module is used to estimate the background zero-sequence admittance and background current analytical signal of each feeder based on the background window, and to subtract the background current analytical signal corresponding to each feeder from the zero-sequence current analytical signal at the beginning of each feeder to obtain the fault residual current analytical signal. The trajectory segmentation module is used to construct the fault equivalent admittance trajectory of each feeder based on the fault residual current analytical signal and the zero-sequence voltage analytical signal, and input the fault equivalent admittance trajectory into the time-series segmentation model to obtain the fault initiation morphology weight. The slope processing module is used to generate an enhanced conductance logarithmic value from the real part of the fault equivalent admittance trajectory at each preset time scale, extract the arc growth slope according to the fault initiation morphology weight, obtain the resonance damping coefficient, and obtain the damped normalized arc growth slope based on the arc growth slope and the resonance damping coefficient. The model discrimination module is used to obtain the model competition discrimination value based on the fitting results of the arc growth model and the normal shear damping model; The result output module is used to extract frequency band energy dispersion and morphological order consistency variables based on the fault equivalent admittance trajectory, and generate detection results by combining the damped normalized arc growth slope and the model competition discriminant.

[0008] Furthermore, to achieve the above objectives, this application also proposes a high-resistance grounding fault detection device based on fault equivalent admittance trajectory. The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program is configured to implement the steps of the high-resistance grounding fault detection method based on fault equivalent admittance trajectory as described above.

[0009] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the high-resistivity grounding fault detection method based on fault equivalent admittance trajectory as described above.

[0010] One or more technical solutions proposed in this application have at least the following technical effects: By acquiring zero-sequence voltage and zero-sequence current at the beginning of each feeder, dividing the detection window, and constructing corresponding analytical signals, the amplitude and phase relationships of each electrical quantity within the detection window can be preserved, providing a data foundation for the synchronous analysis of each feeder. Based on the background window, the background zero-sequence admittance and background current analytical signals of each feeder are estimated. The background current analytical signal is subtracted from the corresponding feeder-end zero-sequence current analytical signal to obtain the fault residual current analytical signal, which can reduce the interference of background zero-sequence current on fault characteristics under normal operating conditions. Based on the fault residual current analytical signal and zero-sequence voltage analytical signal, the fault equivalent admittance trajectory is constructed, and the fault initiation morphology weight is obtained using a time-series segmentation model, which can characterize the time-series changes of the equivalent parameters of the fault channel and highlight the fault initiation interval. In different... The system generates enhanced conductance logarithmic values ​​under a preset time scale, extracts the arc growth slope according to the fault initiation morphology weight, and then uses the resonant damping coefficient for compensation and normalization, which can reduce the influence of arc duration and system damping differences on growth characteristics. Based on the fitting results of the arc growth model and the normal switching damping model, a model competition discriminant is obtained, which can determine whether the current admittance change is more consistent with the arc growth trend or the damping decay trend of normal switching. Frequency band energy dispersion and morphological sequence consistency variables are extracted from the fault equivalent admittance trajectory, and combined with the damped normalized arc growth slope and the model competition discriminant to generate detection results. This allows for judgment from multiple aspects such as frequency band energy distribution, morphological change sequence, and transient evolution trend, reducing misjudgments caused by single features. This application can distinguish between high-resistance grounding faults and normal switching transients based on the bus zero-sequence voltage and the zero-sequence current at the beginning of each feeder. Attached Figure Description

[0011] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1This is a flowchart illustrating the first embodiment of the high-resistivity grounding fault detection method based on fault equivalent admittance trajectory provided in this application. Figure 2 This is a flowchart illustrating the second embodiment of the high-resistivity grounding fault detection method based on fault equivalent admittance trajectory provided in this application. Figure 3 This is a schematic diagram of the module structure of the high-resistivity grounding fault detection device based on the fault equivalent admittance trajectory according to an embodiment of this application; Figure 4 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the high-resistance grounding fault detection method based on fault equivalent admittance trajectory in the embodiments of this application.

[0014] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0015] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0016] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0017] It should be noted that the implementing entity in this application embodiment can be a fault detection system deployed at a substation, switching station, distribution master station, or feeder monitoring site, such as a distribution automation terminal, feeder monitoring terminal, fault recording device, industrial control computer, or distribution master station server. The following description uses a fault detection system as an example.

[0018] Based on this, the first embodiment of this application provides a high-resistivity grounding fault detection method based on fault equivalent admittance trajectory, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the high-resistivity grounding fault detection method based on fault equivalent admittance trajectory in this application.

[0019] In this embodiment, the high-resistivity grounding fault detection method based on fault equivalent admittance trajectory includes steps S10~S60: Step S10: Obtain the zero-sequence voltage and the zero-sequence current at the beginning of each feeder, divide the detection window, and construct the zero-sequence voltage analytical signal and the zero-sequence current analytical signal at the beginning of each feeder. It should be noted that the detection window refers to the data interval to be analyzed extracted from continuously sampled data according to a preset window length. The zero-sequence voltage analytical signal refers to a complex signal with the discrete sequence of zero-sequence voltage as its real part and the corresponding orthogonal component as its imaginary part. The zero-sequence current analytical signal at the feeder head end refers to a complex signal with the discrete sequence of zero-sequence current at the feeder head end as its real part and the corresponding orthogonal component as its imaginary part.

[0020] Understandably, zero-sequence voltage is collected at the zero-sequence voltage output terminal of the distribution network bus voltage transformer, and zero-sequence current is collected at the beginning of each feeder near the bus. The zero-sequence current can be collected by a zero-sequence current transformer or synthesized from synchronously collected three-phase currents. Each current acquisition channel corresponds one-to-one with the feeder identification. The zero-sequence voltage and the zero-sequence current at the beginning of each feeder use the same sampling clock and the sampling time is recorded. Continuous sampling data is extracted according to a preset window length, and adjacent detection windows move sequentially according to a preset step size. The preset window length and preset step size can be determined based on the sampling frequency, power frequency period, and the duration of the transient state to be detected. Orthogonal components are generated for the discrete sequences of zero-sequence voltage and zero-sequence current at the beginning of each feeder within the detection window, forming corresponding analytical signals. This step, by synchronously collecting the electrical quantities at the bus and the beginning of each feeder and establishing a correspondence according to the feeder identification and sampling time, provides a time-consistent data basis for subsequent comparison of the fault characteristics of each feeder.

[0021] Step S20: Estimate the background zero-sequence admittance and background current analytical signal of each feeder based on the background window, and subtract the background current analytical signal corresponding to each feeder from the zero-sequence current analytical signal at the beginning of each feeder to obtain the fault residual current analytical signal. It should be noted that the background window refers to the data interval preceding the current detection window, used to characterize the normal operating state of the distribution network. Background zero-sequence admittance refers to the complex proportional relationship between the zero-sequence current at the feeder's head end and the zero-sequence voltage at the busbar under normal operating conditions. Background current analytical signal refers to the normal zero-sequence current component of the feeder calculated from the background zero-sequence admittance and the current zero-sequence voltage analytical signal. Fault residual current analytical signal refers to the remaining component after subtracting the background current analytical signal from the zero-sequence current analytical signal at the feeder's head end.

[0022] Understandably, for each feeder, the zero-sequence voltage analytical signal and the zero-sequence current analytical signal at the feeder's head end within the background window are read according to the feeder identifier. A complex proportionality coefficient is estimated to minimize the overall deviation between the calculated background current value and the actual sampled value, and this complex proportionality coefficient is used as the background zero-sequence admittance for the corresponding feeder. The background zero-sequence admittance is multiplied by the zero-sequence voltage analytical signal at each sampling point within the current detection window to obtain the background current analytical signal for the corresponding feeder. Then, the background current analytical signal is subtracted from the zero-sequence current analytical signal at the feeder's head end at the same sampling point on the same feeder to obtain the fault residual current analytical signal. This step reduces the impact of inherent feeder-to-ground capacitance, leakage branches, and normal load differences on fault detection by estimating and subtracting the background current component under normal operating conditions for each feeder.

[0023] Step S30: Construct the fault equivalent admittance trajectory of each feeder based on the fault residual current analytical signal and the zero-sequence voltage analytical signal, and input the fault equivalent admittance trajectory into the time-series segmentation model to obtain the fault initiation morphology weight. It should be noted that the fault equivalent admittance trajectory refers to the sequence formed by the change of the complex relationship between the fault residual current analytical signal and the zero-sequence voltage analytical signal over time. The time-series segmentation model refers to a model that divides each sampling point into different transient stages based on the continuous change of the fault equivalent admittance trajectory. The fault initiation mode weight refers to the non-negative weight used to characterize the degree to which each sampling point matches the initiation change mode of a high-resistance grounding fault.

[0024] Understandably, according to the feeder identification, the fault residual current analytical signal of each feeder is paired with the zero-sequence voltage analytical signal at the same sampling time. The complex fault equivalent admittance corresponding to each sampling point is calculated, and these are arranged according to the sampling time to form a fault equivalent admittance trajectory. The real and imaginary changes of the trajectory are extracted and input into a pre-trained time-series segmentation model to obtain the probability that each sampling point is in the admittance rising, nonlinear transition, or falling phase. The weight of the corresponding sampling point is increased based on the probability of the rising phase and the nonlinear transition phase, and the weight of the corresponding sampling point is decreased based on the probability of the falling phase, thus obtaining the fault initiation mode weight. This step, by converting the residual current and zero-sequence voltage into a fault equivalent admittance trajectory that varies with time, can simultaneously retain the conductance and susceptance changes of the fault branch and highlight the fault initiation phase using the time-series change relationship.

[0025] Step S40: At each preset time scale, generate the logarithmic value of enhanced conductance from the real part of the fault equivalent admittance trajectory, extract the arc growth slope according to the fault initiation morphology weight, obtain the resonance damping coefficient, and obtain the damped normalized arc growth slope according to the arc growth slope and the resonance damping coefficient. It should be noted that the preset time scale refers to the length of the time interval used to observe the local changes in the fault equivalent admittance trajectory, which can be selected based on the sampling frequency, power frequency period, and duration of the high-resistance grounding transient. The enhanced conductance logarithmic value refers to the dimensionless characteristic obtained after enhancing, normalizing, and logarithmically transforming the positive growth portion of the real part of the fault equivalent admittance trajectory. The arc growth slope refers to the rate of change of the enhanced conductance logarithmic value over time. The resonant damping coefficient is a non-negative parameter characterizing the decay rate of the zero-sequence resonant component in the distribution network. The damped normalized arc growth slope is a dimensionless discriminant obtained by compensating for and normalizing the arc growth slope using the resonant damping coefficient.

[0026] Understandably, the real part of the fault equivalent admittance trajectory is processed at multiple preset time scales to extract the local change in conductance over time, retaining its non-negative part and enhancing the positive growth component. The processed conductance value is divided by the reference conductance and the logarithm is taken to obtain the enhanced conductance logarithm value for each time scale. The reference conductance can be determined based on the feeder conductance statistics during normal operation or the conductance range of the measurement system. The trend of the enhanced conductance logarithm value and the sampling time is fitted according to the fault initiation morphology weight, so that the sampling point in the fault initiation stage has a greater influence on the fitting result, thus obtaining the arc growth slope. The resonant damping coefficient can be obtained from the decay rate of the zero-sequence voltage transient component or from the existing zero-sequence parameters of the distribution network. The arc growth slope is added to the damping compensation amount, and then normalized using a normalization term including the resonant damping coefficient to obtain the damped normalized arc growth slope. This step extracts conductance growth characteristics through multiple time scales, reducing the sensitivity of a single time scale to changes in arc duration, and utilizes the resonant damping coefficient to reduce the impact of different distribution network damping conditions on arc growth discrimination.

[0027] Step S50: Obtain the model competition discriminant based on the fitting results of the arc growth model and the normal shear damping model; It should be noted that the arc growth model is a fitting model used to describe the time-dependent growth trend of the logarithmic value of enhanced conductance during a high-resistance grounding fault. The normal switching damping model is a fitting model used to describe the oscillating decay trend of the logarithmic value of enhanced conductance during normal switching transients. The model competition discriminant is a discriminant obtained by comparing the fit of the arc growth model and the normal switching damping model to the same set of observation data.

[0028] Understandably, by inputting the sampling times and logarithmic values ​​of enhanced conductance within the same feeder and detection window into the arc growth model and the normal switching damping model respectively, and adjusting the parameters of the two models, the overall deviation between the output values ​​of each model and the logarithmic values ​​of enhanced conductance is reduced. The fitting errors of the two models are compared, and the fitting results are corrected by considering the number of subsequent peak points and the number of model parameters, avoiding bias solely due to differences in the number of parameters. When the correction error of the normal switching damping model is greater than that of the arc growth model, the model competition discriminant is biased towards supporting the arc growth characteristic; conversely, the model competition discriminant is biased towards supporting the normal switching transient characteristic. This step uses two transient evolution models to competitively fit the same set of admittance change data, using the temporal change trend of subsequent peak points to distinguish between arc growth and switching oscillation decay, reducing confusion caused by judging solely based on local amplitudes.

[0029] Step S60: Extract the frequency band energy dispersion and morphological order consistency variables based on the fault equivalent admittance trajectory, and generate the detection result by combining the damped normalized arc growth slope and the model competition discriminant.

[0030] It should be noted that the test results can be the high-resistance grounding fault judgment information corresponding to each feeder, or they can include the candidate fault feeder identifier and the corresponding judgment value.

[0031] Understandably, for the i-th feeder, the frequency band energy dispersion is calculated based on the energy distribution of the real part of the fault equivalent admittance trajectory within multiple preset frequency bands. The continuous growth segment, nonlinear transition segment, and continuous decline segment are determined based on the first and second differences of the smoothed non-negative fault equivalent conductance. A morphological order consistency variable is obtained based on the order of occurrence of each stage. The non-negative parts of the damped normalized arc growth slope under each effective preset time scale are weighted and summed. The weighted summation result, model competition discriminant, frequency band energy dispersion, and morphological order consistency variable are input into a preset logic function to obtain a comprehensive fault detection score. The maximum value is selected from the damped normalized arc growth slope corresponding to each effective preset time scale to obtain the maximum value across preset time scales. When the comprehensive fault detection score, the maximum value across preset time scales, and the model competition discriminant satisfy their respective thresholds, and the morphological order consistency variable is a preset consistent value, the corresponding feeder is added to the candidate feeder set, and a detection result is generated based on the candidate feeder set. This step comprehensively utilizes the arc growth trend, model fitting relationship, frequency band energy distribution, and morphological change sequence to reduce misjudgments caused by a single feature exceeding the threshold.

[0032] This embodiment synchronously collects the bus zero-sequence voltage and the zero-sequence current at the beginning of each feeder. First, it subtracts the background current component under normal operating conditions from the feeder, and then constructs the fault equivalent admittance trajectory. It extracts discrimination information from multiple aspects such as fault initiation mode, conductance growth trend, model fitting relationship, frequency band energy distribution, and morphological change sequence. At the same time, it uses multi-time scale processing and resonant damping normalization to reduce the impact of transient duration and system damping differences, which can improve the ability to distinguish between high-resistance grounding faults and normal switching transients, and provide a basis for determining candidate fault feeders.

[0033] As an example, the steps of acquiring the zero-sequence voltage and the zero-sequence current at the beginning of each feeder, dividing the detection window, and constructing the zero-sequence voltage analytical signal and the zero-sequence current analytical signal at the beginning of each feeder include: synchronously sampling the zero-sequence voltage and the zero-sequence current at the beginning of each feeder to obtain a discrete sequence of zero-sequence voltage and a discrete sequence of zero-sequence current at the beginning of each feeder; dividing the discrete sequence of zero-sequence voltage and the discrete sequence of zero-sequence current at the beginning of each feeder according to a preset window length and a preset sliding step size to obtain a detection window; performing a Hilbert transform on the discrete sequence of zero-sequence voltage located within the detection window to obtain the zero-sequence voltage analytical signal; and performing a Hilbert transform on the discrete sequence of zero-sequence current at the beginning of each feeder located within the detection window to obtain the zero-sequence current analytical signal at the beginning of each feeder.

[0034] It should be noted that the preset window length refers to the number of consecutive sampling points contained in a detection window, which can be determined based on the power frequency cycle and the duration of the transient state to be detected; the preset sliding step size refers to the interval between the starting sampling points of adjacent detection windows, which can be determined based on the detection response speed and the degree of overlap between adjacent detection windows; the boundary extension data refers to the data added to both ends of the detection window before the Hilbert transform to reduce the impact of window truncation.

[0035] Understandably, firstly, a zero-sequence voltage sampling channel is set at the zero-sequence voltage output terminal of the distribution network bus voltage transformer, and a zero-sequence current sampling channel is set at the beginning of each feeder near the bus. Each sampling channel is connected to the same data acquisition device, sharing a sampling clock and sampling trigger signal. Each trigger synchronously reads the zero-sequence voltage and the zero-sequence current at the beginning of each feeder, and configures the same sampling point number according to the sampling order. Each zero-sequence current sampling channel is bound to a corresponding feeder identifier, ensuring that the zero-sequence voltage sampling value under the same sampling point number corresponds to the zero-sequence current sampling value at the beginning of each feeder.

[0036] The synchronous sampling frequency can be set from 5 kHz to 20 kHz, depending on the transient frequency range to be retained and the sampling capability of the data acquisition equipment. For example, for a distribution network with a power frequency of 50 Hz, the synchronous sampling frequency can be set to 10 kHz, so that each power frequency cycle contains 200 sampling points. The sampled data are arranged sequentially according to the sampling point number, forming a discrete sequence of zero-sequence voltage and a discrete sequence of zero-sequence current at the feeder head corresponding to each feeder identifier.

[0037] Discrete sequences are synchronously extracted according to a preset window length and a preset sliding step size, ensuring that the zero-sequence voltage discrete sequence and the zero-sequence current discrete sequence at the beginning of each feeder have the same starting sampling point, ending sampling point, and number of sampling points within the same detection window. The preset window length can be 1 to 4 power frequency cycles, and the preset sliding step size can be 1 / 4 to 1 / 2 of the preset window length. For example, when the sampling frequency is 10 kHz, the preset window length can be set to 400 sampling points, corresponding to 2 power frequency cycles, and the preset sliding step size can be set to 100 sampling points, corresponding to half a power frequency cycle. If the sampled data is insufficient for a complete detection window, sampling data continues to be received until the number of sampling points reaches the preset window length before a detection window is formed, without using zero-padding to generate an incomplete detection window.

[0038] Based on this, boundary extensions are performed on the discrete sequences of zero-sequence voltage within the detection window and the discrete sequences of zero-sequence current at the beginning of each feeder. The boundary extension length can be half to one power frequency cycle. For example, 200 sampling points are added at both ends of the detection window in a mirror manner. During mirror extension, the sampling values ​​near the beginning of the detection window are added in reverse order before the beginning, and the sampling values ​​near the end are added in reverse order after the end, to avoid abrupt amplitude changes at the window boundaries caused by zero padding.

[0039] A Discrete Fourier Transform (DFT) is performed on the zero-sequence voltage discrete sequence after boundary extension. The DC component is retained in the frequency domain, the positive frequency components are doubled, and the negative frequency components are set to zero. An Inverse DFT is then performed to obtain the complex sequence corresponding to the extended interval. The complex sequence corresponding to the sampling point range of the original detection window is truncated to obtain the zero-sequence voltage analytical signal. The same method is used to process the discrete sequences of the zero-sequence current at the beginning of each feeder to obtain the zero-sequence current analytical signal at the beginning of each feeder. The truncated analytical signals retain the original sampling point number and feeder identifier, thus maintaining the time correspondence between the zero-sequence voltage analytical signal and the zero-sequence current analytical signal at the beginning of each feeder.

[0040] This example ensures that the bus zero-sequence voltage and the zero-sequence current at the beginning of each feeder are consistent at the sampling time by sharing a sampling clock and sampling trigger signal; it avoids phase deviation caused by window misalignment of different electrical quantities by synchronously dividing detection windows with the same boundary; and it reduces the boundary distortion at both ends of the window by performing Hilbert transform after mirror extension, providing a complex signal with a clear correspondence for subsequent background zero-sequence admittance estimation and fault equivalent admittance trajectory construction.

[0041] As an example, the step of estimating the background zero-sequence admittance and background current analytical signal of each feeder based on the background window, and subtracting the background current analytical signal corresponding to each feeder from the zero-sequence current analytical signal at the beginning of each feeder to obtain the fault residual current analytical signal includes: selecting continuous sampling points before the detection window to obtain a background window; calculating candidate background zero-sequence admittance for each feeder based on the zero-sequence current analytical signal at the beginning of each feeder and the zero-sequence voltage analytical signal; updating or freezing the background zero-sequence admittance of each feeder based on the average energy of the zero-sequence voltage analytical signal within the background window, the change of the candidate background zero-sequence admittance relative to the previous effective background zero-sequence admittance, and the detection result of the previous detection window containing the background window; obtaining the background current analytical signal based on the background zero-sequence admittance and the zero-sequence voltage analytical signal, and subtracting the background current analytical signal corresponding to each feeder from the zero-sequence current analytical signal at the beginning of each feeder to obtain the fault residual current analytical signal.

[0042] It should be noted that candidate background zero-sequence admittance refers to the complex admittance estimated based on the zero-sequence voltage analytical signal and the corresponding feeder head zero-sequence current analytical signal within the current background window, which has not yet passed the update condition judgment; average energy refers to the average value of the square of the magnitude of the zero-sequence voltage analytical signal within the background window; admittance change refers to the degree of relative change of candidate background zero-sequence admittance relative to the previous effective background zero-sequence admittance; the previous effective background zero-sequence admittance refers to the background zero-sequence admittance that most recently met the background zero-sequence admittance update condition and was saved; updating background zero-sequence admittance means replacing the previous effective background zero-sequence admittance with candidate background zero-sequence admittance; freezing background zero-sequence admittance means stopping the use of candidate background zero-sequence admittance and continuing to use the previous effective background zero-sequence admittance.

[0043] In this example, the detection result from the previous detection window is used to determine the detection status of the corresponding feeder. If the detection result does not indicate a high-resistance grounding fault risk on the corresponding feeder, the feeder is in normal condition, and the feeder status flag is 0. If the detection result indicates a high-resistance grounding fault risk on the corresponding feeder, the feeder status flag is 1, and the background zero-sequence admittance update for the corresponding feeder is frozen. High-resistance grounding fault risk includes suspected fault status or fault alarm status; the background zero-sequence admittance update judgment does not require further differentiation between these two status levels.

[0044] Understandably, firstly, a background window is selected from consecutive sampling points within the time range of the previous detection window, preceding the starting sampling point of the current detection window. This background window does not overlap with the current detection window. The number of sampling points contained in the detection window is denoted as... The number of sampling points contained in the background window is denoted as ,in, The background window length is determined based on the stabilization time of the zero-sequence voltage under normal operating conditions, the sampling frequency, and the length of available data in the previous detection window. If the effective data before the current detection window is insufficient to meet the preset background window length, data continues to accumulate, and candidate background zero-sequence admittance is not estimated at this time.

[0045] For the i-th feeder, read the zero-sequence voltage analytical signal and the zero-sequence current analytical signal at the feeder head end corresponding to the sampling time within the background window, and calculate the candidate background zero-sequence admittance according to the following formula: In the formula, i is the feeder number and n is the sampling point number. The set of sampling point numbers corresponding to the background window. The candidate background zero-order admittance for the i-th feed line. Let be the value of the zero-sequence current analytical signal at the beginning of the i-th feeder at the n-th sampling point. The value of the zero-sequence voltage analytical signal at the nth sampling point. This is a complex conjugate operation. The preset voltage square stability term is positive. The sum of squared voltage magnitudes in the denominator is... The dimensions of all are V 2 The dimensions of the candidate background zero-sequence admittance are Siemens. The preset voltage square stability term can be taken as 1 to 10 times the sum of squares of the noise voltage magnitudes within the background window to avoid large fluctuations in the admittance estimation results when the zero-sequence voltage amplitude is small.

[0046] Next, the average energy of the zero-sequence voltage analytical signal within the background window is calculated according to the following formula: In the formula, The average energy of the zero-sequence voltage analytical signal within the background window. This represents the number of sampling points contained in the background window. The units for both the average energy and the preset energy threshold are V. 2 The preset energy threshold can be set to 4 to 25 times the mean square value of the noise in the zero-sequence voltage sampling channel, depending on the noise level and quantization accuracy of the sampling channel, so that the zero-sequence voltage involved in admittance estimation is higher than the measurement noise.

[0047] For the i-th feedline that has already been initialized, the change in the candidate background zero-sequence admittance relative to the previous effective background zero-sequence admittance is calculated according to the following formula: In the formula, This is a dimensionless change in admittance. Let the previous effective background zero-sequence admittance be the i-th feed line. The preset admittance stability term is positive, with dimensions in Siemens. It can be taken as the ratio of the minimum current that the zero-sequence current sampling channel can resolve to the minimum voltage that the zero-sequence voltage sampling channel can resolve. The preset change threshold can be set to 0.05~0.2, specifically determined based on the fluctuation range of the background zero-sequence admittance during normal operation.

[0048] Before fault detection is implemented, 3 to 5 background windows can be continuously acquired during operation periods when there are no grounding faults and no switching operations in the distribution network, and candidate background zero-sequence admittances can be calculated for each window. When the changes in adjacent candidate background zero-sequence admittances are not higher than a preset change threshold, the average value of each candidate background zero-sequence admittance is taken as the initial previous valid background zero-sequence admittance. If initialization data cannot be obtained, the most recently saved valid background zero-sequence admittance under the same distribution network connection mode can be read; if no available historical value exists, background data acquisition continues, and the calculation of fault residual current analytical signal is not started temporarily.

[0049] After initialization, for the i-th feeder, when the average energy reaches the preset energy threshold, the change in the candidate background zero-sequence admittance relative to the previous effective background zero-sequence admittance is not higher than the preset change threshold, and the detection result of the previous detection window containing the background window does not indicate that the i-th feeder has a high-resistance grounding fault risk, the candidate background zero-sequence admittance is used as the current background zero-sequence admittance of the i-th feeder, and the candidate background zero-sequence admittance is saved as the new previous effective background zero-sequence admittance; if any condition is not met, the background zero-sequence admittance update of the i-th feeder is frozen, and the original previous effective background zero-sequence admittance is used as the current background zero-sequence admittance.

[0050] The corresponding update relationships are as follows: In the formula, The background zero-sequence admittance currently used for the i-th feeder is... To preset the energy threshold, The average energy of the zero-sequence voltage analytical signal within the background window. To preset the change threshold, For candidate background zero-order admittance, For the previous effective background zero-sequence admittance, This is the status flag for the i-th feeder determined based on the detection results of the previous detection window, which includes the background window; if the detection results do not indicate a high-resistance grounding fault risk in the i-th feeder, When the test results indicate that the i-th feeder has a high-resistance grounding fault risk, When the formula uses the first value, the candidate background zero-order admittance is saved as the previous valid background zero-order admittance used in the next detection window; when the second value is used, the previous valid background zero-order admittance remains unchanged.

[0051] Finally, the background zero-sequence admittance currently used for each feeder is multiplied by the zero-sequence voltage analytical signal within the current detection window to obtain the background current analytical signal. Then, the background current analytical signal is subtracted from the zero-sequence current analytical signal at the feeder head with the same sampling time and feeder identification to obtain the fault residual current analytical signal. In the formula, The background current analytical signal value of the i-th feeder at the n-th sampling point is obtained. The value of the fault residual current analytical signal of the i-th feeder at the n-th sampling point is taken.

[0052] This example estimates the zero-sequence admittance of each feeder under normal operating conditions by setting a background window separate from the current detection window. It uses average energy, admittance change, and the detection results of the previous detection window to control the updating and freezing of the background zero-sequence admittance, reducing the impact of low zero-sequence voltage, measurement noise, and data during suspected faults or fault alarms on the background zero-sequence admittance. Based on this, the background current component of each feeder is subtracted, and the residual current related to the current transient is retained, providing input data with less background influence for the subsequent construction of the fault equivalent admittance trajectory.

[0053] As an example, the steps of generating the enhanced conductance logarithm value from the real part of the fault equivalent admittance trajectory at each preset time scale, extracting the arc growth slope according to the fault initiation morphology weight, obtaining the resonant damping coefficient, and obtaining the damped normalized arc growth slope based on the arc growth slope and the resonant damping coefficient include: extracting the fault equivalent conductance peak value from the real part of the fault equivalent admittance trajectory at each preset time scale, calculating the enhanced conductance logarithm value based on the fault equivalent conductance peak value; and weighting the enhanced conductance logarithm value according to the fault initiation morphology weight. The arc growth slope is obtained by fitting the relationship between the logarithmic value of the enhanced conductance and the sampling time; the normalized logarithmic value of the zero-sequence voltage envelope is obtained by dividing the envelope of the zero-sequence voltage analytical signal by the reference voltage and taking the logarithm; the inverse of the fitting slope of the normalized logarithmic value of the zero-sequence voltage envelope relative to the sampling time is taken, and the larger value is compared with zero, or the resonant damping coefficient is obtained by calculating based on the equivalent zero-sequence damping resistance and the equivalent zero-sequence capacitance of the system in parallel; the damped normalized arc growth slope is calculated based on the arc growth slope and the resonant damping coefficient.

[0054] It should be noted that the fault equivalent conductance peak value refers to the local maximum value of the real part of the fault equivalent admittance trajectory within a preset time scale; the positive growth of adjacent peak values ​​refers to the non-negative difference between the current fault equivalent conductance peak value and the previous fault equivalent conductance peak value; the reference conductance refers to the positive conductance used to convert the fault equivalent conductance peak value into a dimensionless ratio; and the reference voltage refers to the positive voltage used to convert the zero-sequence voltage envelope into a dimensionless ratio.

[0055] Understandably, the process begins by extracting the real part of the fault equivalent admittance trajectory of the i-th feeder to obtain the fault equivalent conductance sequence, and then searching for local maxima at multiple preset time scales. The preset time scales can be half a power frequency cycle, one power frequency cycle, or two power frequency cycles, or they can be determined based on the pulse interval of the high-resistance grounding arc, the sampling frequency, and the detection window length. For example, in a distribution network with a power frequency of 50Hz, the aforementioned preset time scales are approximately 10ms, 20ms, and 40ms, respectively. For any preset time scale, the corresponding time length is used as the peak search interval, and the maximum fault equivalent conductance value within the search interval is taken as the fault equivalent conductance peak value. When the interval between adjacent peaks is less than the current preset time scale, only the peak with the larger value is retained; when the peak values ​​are the same, the peak with the earlier sampling time is retained. When the search intervals at both ends of the detection window exceed the window boundaries, only the existing fault equivalent conductance values ​​within the window are compared.

[0056] Arrange the peak values ​​of the fault equivalent conductance at the m-th preset time scale according to the sampling time, and calculate the logarithmic value of the enhanced conductance according to the following formula: In the formula, m is the preset time scale number, and k is the fault equivalent conductance peak value number. Let $T$ be the peak value of the equivalent conductance of the $k$-th fault of the $i$-th feeder at the $m$-th preset time scale. for The non-negative part, This represents the positive increase in the peak equivalent conductance of the k-th fault relative to the peak equivalent conductance of the (k-1)-th fault at the m-th preset time scale for the i-th feeder. The preset difference enhancement coefficient is non-negative. As a positive reference conductance, The pre-defined logarithmic stable term is positive. This is the logarithmic value of the enhanced conductance corresponding to the peak equivalent conductance of the k-th fault at the m-th preset time scale for the i-th feeder. For the first peak equivalent conductance of the fault, the corresponding positive increase is set to 0.

[0057] , and All dimensions are Siemens. , and All are dimensionless quantities. The preset differential enhancement coefficient can be set to 0~2, where 0 indicates no enhancement of the positive growth of adjacent peak values. The specific value can be determined based on the difference in conductance growth between high-resistance grounding fault samples and normal switching samples. The reference conductance can be the median of the positive peak value of the fault equivalent conductance in the normal operation data, or it can be the ratio of the reference current to the reference voltage. The preset logarithmic stability term can be set to 10. -6 ~10 -3 Specifically, it is determined based on the ratio of the conductivity measurement resolution to the reference conductivity.

[0058] Next, for the m-th preset time scale of the i-th feeder, the fault equivalent conductance peak values ​​that are retained according to the peak search rule and whose fault initiation mode weight is greater than 0 are called positive weight fault equivalent conductance peak values, and their peak indices are used to form a slope fitting peak index set. The sampling point number corresponding to the k-th fault equivalent conductance peak value is denoted as . The corresponding sampling time is recorded as ,in To synchronize the sampling frequency and read the sampling points Fault initiation mode weight The weighted time mean, the weighted logarithmic mean of enhanced conductance, and the arc growth slope are calculated using the following formulas: In the formula, This is the weighted time mean at the m-th preset time scale, in seconds; To calculate the weighted average logarithmic conductance; The slope of the electric arc growth is the reciprocal of time. The term is a positively stable term with dimensions in seconds squared, which can be determined based on the synchronous sampling interval, for example, by making... No more than 1% of the square of the synchronous sampling interval.

[0059] when If there are fewer than two peak numbers, the arc growth slope corresponding to the preset time scale will be recorded as invalid. The logarithmic value of the enhanced conductance is a dimensionless quantity, and the unit of sampling time is seconds. Therefore, the dimension of the arc growth slope is the reciprocal of time.

[0060] For a zero-sequence voltage analytical signal, the complex amplitude of each sampling point is taken as the zero-sequence voltage envelope, and the normalized logarithmic value of the zero-sequence voltage envelope is calculated according to the following formula: In the formula, Let be the normalized logarithmic value of the zero-sequence voltage envelope at the nth sampling point. The reference voltage is positive. It can be the maximum value of the zero-sequence voltage envelope within the detection window, or the rated voltage of the zero-sequence voltage sampling channel. A fixed reference voltage is used within the same detection window; therefore, the choice of reference voltage affects the fitting intercept but does not change the envelope attenuation slope. The zero-sequence voltage envelope used in the calculation should be higher than the noise amplitude of the sampling channel to avoid taking the logarithm of zero.

[0061] In one alternative implementation, for the i-th feeder, the sampling point where the first occurrence of the maximum value of the fault initiation mode weight within the detection window is denoted as... ,from Then, sampling points whose probability during the descent phase reaches a preset descent phase probability threshold are selected to form a set of descent phase sampling points: In the formula, Let i be the set of sampling points for the descent phase corresponding to the i-th feeder. Let be the probability of the i-th feeder during the descent phase at the n-th sampling point. The preset probability threshold for the descent phase can be selected within the range of 0.5 to 0.8.

[0062] When the number of sampling points in the sampling point set during the descent phase reaches the preset minimum number of sampling points, the descent phase probability is used as the fitting weight, and the weighted time mean and the weighted zero-sequence voltage envelope normalized logarithmic mean are calculated according to the following formulas: In the formula, Let n be the sampling time of the nth sampling point. This represents the weighted time mean of the descent phase corresponding to the i-th feeder. Let be the normalized logarithmic mean of the weighted zero-sequence voltage envelope during the falling phase corresponding to the i-th feeder.

[0063] The fitting slope of the normalized logarithm of the zero-sequence voltage envelope relative to the sampling time is calculated based on the weighted time mean and the weighted normalized logarithm mean of the zero-sequence voltage envelope: By taking the inverse of the fitted slope and comparing it with 0, we obtain the resonant damping coefficient estimated based on the zero-sequence voltage envelope decay process: In the formula, For the set of sampling points in the falling phase corresponding to the i-th feeder, the normalized logarithmic value of the zero-sequence voltage envelope relative to the sampling time is the fitting slope. The resonant damping coefficient is estimated based on the descent phase. The term is a positively stable term with dimensions in seconds squared, which can be determined based on the synchronous sampling interval, for example, by making... No more than 1% of the square of the synchronous sampling interval. and The dimensions of all of them are the reciprocals of time.

[0064] The preset minimum number of sampling points is determined based on the synchronous sampling frequency and the shortest duration that the descent phase needs to cover. For example, if the descent phase is required to cover at least half a power frequency cycle, the preset minimum number of sampling points can be set according to the number of sampling points contained in half a power frequency cycle.

[0065] When the number of sampling points during the descent phase does not reach the preset minimum number of sampling points, the resonant damping coefficient based on the system's zero-sequence parameters is calculated according to the system's parallel equivalent zero-sequence damping resistance and the system's equivalent zero-sequence capacitance. In the formula, The resonant damping coefficient is calculated based on the system's zero-sequence parameters. The system is connected in parallel with an equivalent zero-sequence damping resistance, in ohms. The system's equivalent zero-sequence capacitance has the dimension of farad. The dimension of is the reciprocal of time. The system's parallel equivalent zero-sequence damping resistance and system equivalent zero-sequence capacitance are pre-configured according to the current distribution network wiring method before being put into fault detection.

[0066] The resonance damping coefficient used in the current detection window of the i-th feeder is determined according to the following relationship: In the formula, The set of sampling points for the descent phase of the i-th feeder Number of sampling points included A minimum number of sampling points is preset. When the number of sampling points during the descent phase reaches the preset minimum number, the zero-sequence voltage envelope fitting result corresponding to the current detection window is used; when the number of sampling points during the descent phase is insufficient, the system zero-sequence parameter calculation result corresponding to the current distribution network connection method is used. When the system zero-sequence parameters are not pre-configured and the number of sampling points during the descent phase of the i-th feeder is insufficient, the resonant damping coefficient corresponding to the i-th feeder in the current detection window is marked as invalid, and the i-th feeder is not included in the candidate feeder judgment of the current detection window.

[0067] Finally, for each preset time scale, the corresponding arc growth slope and resonance damping coefficient are substituted into the following formula to obtain the damped normalized arc growth slope: In the formula, Let be the damped, normalized arc growth slope of the i-th feeder at the m-th preset time scale. Let be the arc growth slope of the i-th feeder at the m-th preset time scale. λ is a non-negative preset damping compensation coefficient. The subscript λ indicates that this parameter is related to the resonant damping coefficient and does not represent another variable. The preset normalization stability term is positive, used to avoid abnormally large normalization results when the resonant damping coefficient is close to zero. It can be set to 0~1, and the specific value is determined according to the range of change of the arc growth slope under different resonant damping conditions; It can be taken as 1% to 10% of the median of the historical effective resonance damping coefficient. , and The dimensions of all of them are the reciprocals of time. and All are dimensionless quantities. When historical resonance damping data is lacking, the detection window duration can be used as a reference. Determine the preset normalized stable term: in, It is a dimensionless coefficient and can be selected in the range of 0.01 to 0.1.

[0068] This example extracts the peak equivalent conductance of the fault at multiple time scales and enhances the positive growth component of adjacent peaks, thus preserving the arc growth characteristics under different durations. By weighting the fault initiation morphology, the influence of the peak value in the fault initiation stage on the arc growth slope is improved. Furthermore, the arc growth slope is compensated and normalized using the resonant damping coefficient, reducing the impact of different distribution network damping conditions on the slope discrimination results. This provides a dimensionless discrimination quantity for distinguishing between high-resistance grounding faults and normal switching transients.

[0069] As an example, the step of obtaining the model competition discriminant based on the fitting results of the arc growth model and the normal shear damping model includes: taking the sampling point where the maximum value of the fault initiation morphology weight first appears within the detection window as the starting point, taking the sampling points corresponding to the fault equivalent conductance peak value after the starting point as subsequent peak points, and obtaining the relative time of the subsequent peak points relative to the starting point and the number of subsequent peak points; calculating the output value of the arc growth model and the output value of the normal shear damping model based on the relative time and the model parameters to be fitted; when the number of subsequent peak points meets the preset fitting conditions, fitting the arc growth model and the normal shear damping model respectively according to the fault initiation morphology weight to obtain two types of model errors, and obtaining the model competition discriminant based on the two types of model errors and the corresponding number of model parameters.

[0070] It should be noted that subsequent peak points refer to the sampling points located after the first occurrence of the maximum value of the fault initiation mode weight, corresponding to the peak value of the fault equivalent conductance; relative time refers to the difference between the sampling time of subsequent peak points and the sampling time of the starting point; correction error refers to the fitting error obtained after correcting the sum of squared errors using the number of subsequent peak points and the number of model parameters; preset fitting condition refers to the number of subsequent peak points being greater than the larger of the number of parameters of the two types of models; preset neutral value refers to the model competition discriminant used when the number of subsequent peak points is insufficient or both types of models fail to fit. In this example, the preset neutral value is 0.

[0071] Understandably, for the i-th feeder, the maximum value of the fault initiation morphology weight is found within the detection window, and the sampling point where this maximum value first appears is recorded as the starting point. . This is the set of valid preset time scale indices for participating in model competition fitting. The set of fitting peak indices is composed of the peak indices at the m-th preset time scale that are after the starting point and have a fault initiation morphology weight greater than 0. .

[0072] The time difference between the k-th fitted peak point and the starting point at the m-th preset time scale is: In the formula, Let be the sampling point number corresponding to the k-th fitted peak value of the i-th feed line at the m-th preset time scale. To synchronize the sampling frequency, The unit is seconds.

[0073] The output values ​​of the arc growth model and the normal shear damping model are calculated based on the relative time and the parameters of the model to be fitted: In the formula, The output value for the arc growth model. This is the output value of the normal shear damping model. This is the parameter set for the arc growth model. , , These are the bias term, growth magnitude, and growth rate coefficient, respectively. This is the parameter set for the normal shear damping model. , , , and These are the bias term, oscillation amplitude, attenuation coefficient, damping angular frequency, and initial phase, respectively.

[0074] , , , All are dimensionless quantities. , , The dimensions of all of them are the reciprocals of time. It is a dimensionless quantity. During fitting, restrictions are applied. , , , , and will The time constant is limited to [-π, π]. Based on the shortest time constant of the transient state to be detected. and the longest time constant Set the value ranges for the growth rate coefficient and attenuation coefficient, based on the lower limit of the zero-sequence resonant frequency of the distribution network. and upper limit Set the range of values ​​for the damping angular frequency: and It can be determined based on the transient duration covered by the detection window. and It can be determined based on the zero-sequence parameter calculation results or field test data corresponding to the current power distribution network wiring method.

[0075] The number of subsequent peak points involved in the fitting of the two types of models is: in, This represents the set of fitted peak indices for the i-th feed line at the m-th preset time scale. The number of peak indices contained in this set. This is the set of valid preset time scales for participating in model competition fitting. This represents the number of subsequent peak points that participate in the fitting of the two types of models.

[0076] The arc growth model contains 3 parameters to be fitted, and the normal shear damping model contains 5 parameters to be fitted. When the value is no greater than 5, the two types of models are not fitted. The model competition discriminant of the i-th feeder is set to 0 and used as a neutral input for subsequent calculation of the comprehensive fault detection score.

[0077] when At that time, the logarithmic values ​​of the enhanced conductance under each effective preset time scale are used to jointly fit the two types of models. The fitting parameters of the arc growth model corresponding to the i-th feeder are obtained using the nonlinear weighted least squares method. Fitting parameters with normal shear damping model The sum of squared errors for the two types of models is calculated using the following formula: In the formula, and Let Si represent the sum of squared errors of the arc growth model and the sum of squared errors of the normal switching damping model corresponding to the i-th feeder, respectively. This is the logarithmic value of the enhanced conductance corresponding to the k-th fitted peak value of the i-th feeder at the m-th preset time scale; Weights for the initial fault morphology at the same sampling point; Let be the non-negative scale weight of the i-th feeder at the m-th effective preset time scale. The sum of the scale weights corresponding to each effective preset time scale is 1, i.e. .

[0078] The scale weights can be set to be equal, or determined based on the results of distinguishing between high-resistance grounding faults and normal switching transients at each preset time scale in the validation samples. The logarithmic value of enhanced conductance, the weight of fault initiation mode, and the sum of squared errors of the two types of models are all dimensionless quantities.

[0079] when If the error is greater than 5 and at least one model fits successfully, the correction error is calculated according to the following formula; if any model fails to fit, the corresponding preset penalty value is used as the sum of squared errors of that model: In the formula, and These are the correction errors for the arc growth model and the normal switching damping model corresponding to the i-th feeder, respectively. Let be the number of parameters in the arc growth model, and , The number of parameters in the normal shear damping model, and ; The preset error stability term is positive; it is a dimensionless quantity and can be set to 10. -8 ~10 -4 The specific value is determined based on the numerical resolution of the enhanced conductance logarithm.

[0080] Taking the logarithm of the ratio of the correction error of the normal switching damping model to the correction error of the arc growth model, we obtain the model competition discriminant for the i-th feeder: when When the correction error of the arc growth model is small, the current logarithmic value of enhanced conductance better matches the arc growth trend; when At that time, the correction error of the normal switching damping model is small, and the current logarithmic value of the enhanced conductance is more consistent with the oscillation decay trend; when When the value is close to 0, the fit between the two types of models is similar.

[0081] when When the value is no greater than 5, the model competition discriminant is set to 0 and used as a neutral input in the calculation of the comprehensive fault detection score. If the nonlinear weighted least squares method fails to converge within a preset number of iterations, the fitted parameters exceed the limit range, or the sum of squared errors is not a finite value, the sum of squared errors of the corresponding model is set to a preset penalty value. If only one model fails to fit, the model competition discriminant is still calculated based on the sum of squared errors of both models. If both models fail to fit, the model competition discriminant is set to 0. The preset penalty value is determined based on the distribution of the sum of squared errors of the models in the training samples or historical waveform data, and is greater than the upper limit of the sum of squared errors when the model fits normally.

[0082] This example uses subsequent peak data at various effective preset time scales to jointly fit the arc growth model and the normal switching damping model, and utilizes the time series difference between admittance growth and oscillation decay to form a model competition discriminant. The error correction takes into account the fitting deviation caused by the different number of parameters in the two types of models. When there are insufficient peak points or both types of models fail to fit, a preset neutral value is used, and when a single model fails to fit, a preset penalty value is used, which reduces the impact of unreliable fitting results on high-resistance grounding fault detection.

[0083] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 , Figure 2This is a flowchart illustrating the second embodiment of the high-resistivity grounding fault detection method based on fault equivalent admittance trajectory of this application. Step S30 of the high-resistivity grounding fault detection method based on fault equivalent admittance trajectory includes steps S31 to S34: Step S31: For each feeder, construct the fault equivalent admittance trajectory based on the fault residual current analytical signal and the zero-sequence voltage analytical signal; Step S32: Extract the real and imaginary parts of the fault equivalent admittance trajectory to obtain the fault equivalent conductance sequence and the fault equivalent susceptance sequence; Step S33: Input the fault equivalent conductance sequence and the fault equivalent susceptance sequence into the time series segmentation model to obtain the rising phase probability, the nonlinear transition phase probability and the falling phase probability. Step S34: Calculate the fault initiation mode weight based on the rising phase probability, the nonlinear transition phase probability, and the falling phase probability.

[0084] It should be noted that the voltage square stability term refers to the positive number added to the denominator of the zero-sequence voltage modulus square, used to limit the calculation result of the fault equivalent admittance under low voltage conditions, and its dimension is V. 2 The stage labels include rising stage label, nonlinear transition stage label, linear transition stage label, falling stage label, and background stage label. A rising stage label is configured when the fault equivalent conductance shows an increasing trend within continuous sampling points, and the first-order difference is positive and exceeds the corresponding threshold; a nonlinear transition stage label is configured when the fault equivalent conductance is still changing, but the first-order or second-order difference fluctuates significantly, or the local linear fitting residual exceeds the corresponding threshold; a linear transition stage label is configured when the fault equivalent conductance shows an approximately monotonic change and the local linear fitting residual is below the corresponding threshold; a falling stage label is configured when the fault equivalent conductance shows a decreasing trend, and the first-order difference is negative and exceeds the corresponding threshold; and a background stage label is configured when no significant transient change occurs. Each stage must have at least a preset number of sampling points. The relevant thresholds and the number of continuous sampling points are determined based on training samples, historical waveform data, or on-site settings.

[0085] When configuring stage labels, first determine whether the sampling point meets the configuration conditions for rising or falling stage labels; if the configuration conditions for rising and falling stage labels are not met, then determine whether the configuration conditions for nonlinear transition stage labels are met; if they are still not met, then determine whether the configuration conditions for linear transition stage labels are met; sampling points outside the event interval that do not show obvious transient changes are configured as background stage labels, so that each sampling point corresponds to only one stage label.

[0086] Understandably, firstly, for the i-th feeder, the fault residual current analytical signal at the n-th sampling point is paired with the zero-sequence voltage analytical signal, and the fault equivalent admittance is calculated according to the following formula: In the formula, Let be the fault equivalent admittance of the i-th feeder at the n-th sampling point. Let be the value of the fault residual current analytical signal of the i-th feeder at the n-th sampling point. The value of the zero-sequence voltage analytical signal at the nth sampling point. This is a complex conjugate operation. The modulus of a complex number The voltage square stability term is positive. The fault equivalent admittance within the detection window is arranged according to the sampling time to form the fault equivalent admittance trajectory for the corresponding feeder.

[0087] The molecular weight is A⋅V, and the denominator is V. 2 Therefore, the dimension of the fault equivalent admittance is Siemens. The voltage square stability term can be taken as 1 to 10 times the mean square value of the noise voltage of the zero-sequence voltage sampling channel, depending on the noise level and voltage resolution of the sampling channel.

[0088] Then, the real and imaginary parts of the fault equivalent admittance trajectory are extracted respectively to obtain the fault equivalent conductance sequence and the fault equivalent admittance sequence: In the formula, Let the fault equivalent conductance of the i-th feeder at the n-th sampling point be . The corresponding fault equivalent susceptance, both are in Siemens units. To take the real part of a complex number, This indicates taking the imaginary part of a complex number. The sampling points are arranged in the original sampling order to ensure that the fault equivalent conductance sequence and the fault equivalent susceptance sequence remain in constant correspondence.

[0089] For including A detection window with sampling points is formed, arranging the fault equivalent conductance sequence and fault equivalent susceptance sequence according to the corresponding sampling points, creating a detection window with a size of [size missing]. The model input matrix is ​​given, where the first column represents the fault equivalent conductance and the second column represents the fault equivalent susceptance. Before inputting the data into the model, the input is standardized using the mean and standard deviation of each input channel from the training samples; if the standard deviation of any input channel is 0, the corresponding standard deviation is set to 1.

[0090] The temporal segmentation model employs a one-dimensional U-Net network, which includes an encoder, a bottleneck layer, a decoder, and an output layer. The encoder comprises multiple levels of one-dimensional convolutional modules and downsampling modules. Each level of the one-dimensional convolutional module includes a one-dimensional convolutional layer, a normalization layer, and a non-linear activation layer. Adjacent levels are connected via downsampling modules. The bottleneck layer is connected to the final stage of the encoder. The decoder includes upsampling modules and one-dimensional convolutional modules corresponding to the encoder's levels. Each level of the decoder receives features from the encoder at the same time scale through skip connections. The output layer uses a one-dimensional convolutional layer to map the features corresponding to each sampling point to a 5-dimensional output, which is then processed by a Softmax function to obtain the probabilities of the rising phase, non-linear transition phase, linear transition phase, falling phase, and background phase. The model output size is [size missing]. The nth row of the output matrix corresponds to the same sampling point as the nth row of the input matrix. The probabilities of the five stages corresponding to the same sampling point are all between 0 and 1, and their sum is 1. The linear transition stage probability and the background stage probability are used to help distinguish the transition process and the background state, and are not substituted into the fault initiation form weight formula.

[0091] The training samples include high-resistance grounding fault samples, low-resistance grounding fault samples, load switching samples, capacitor switching samples, distributed generation grid-connected samples, and normal operation samples. These can be derived from historical waveform recordings, experimental platform waveform recordings, and simulation data obtained according to distribution network operating parameters. The training samples use the same synchronous sampling frequency and detection window length as the actual detection process. Each training sample contains... Each set of sampling points is used. Training samples are divided into training and validation sets, based on a single fault event, a normal switching event, a distributed power grid connection event, or a continuous normal operation period. This avoids adjacent detection windows formed by the same event or the same continuous operation period from simultaneously entering the training and validation sets. The training set can account for 70% to 80% of all event samples, and the validation set can account for 20% to 30%.

[0092] Each training sample is labeled with a stage label. The event intervals in the training samples are determined based on the fault waveform records, switching test records, or the start and end times of events in the simulation settings. When the fault equivalent conductance shows a continuous increasing trend within consecutive sampling points, and the first-order difference is positive and exceeds the growth threshold, the corresponding sampling point is labeled as an rising stage. When the fault equivalent conductance is still changing, but the first-order or second-order difference fluctuations exceed the corresponding threshold, or the local linear fitting residual exceeds the residual threshold, the corresponding sampling point is labeled as a nonlinear transition stage. When the fault equivalent conductance shows an approximately monotonic change, and the local linear fitting residual is not higher than the residual threshold, the corresponding sampling point is labeled as a linear transition stage. When the fault equivalent conductance shows a decreasing trend, and the first-order difference is negative and exceeds the decreasing threshold, the corresponding sampling point is labeled as a decreasing stage. Sampling points outside the event interval that do not show significant transient changes are labeled as background stages. Each threshold and the minimum number of continuous sampling points are determined based on the training samples, historical waveform data, or field settings.

[0093] During training, the input matrix is ​​first standardized using the mean and standard deviation of the fault equivalent conductance and susceptance channels in the training set. The standardized input matrix is ​​then input into the time-series segmentation model. A weighted cross-entropy loss is calculated using point-by-point stage labels, and the model parameters are updated via backpropagation. The weights for each class are determined based on the number of sampling points for that class in the training set, with increased weights for the rising, nonlinear, and linear transition stages to mitigate the impact of a small sample size during the short-term fault initiation stage. The optimizer can be the Adam optimizer, with an initial learning rate set to... The batch size can be set to 16~64, and the maximum number of training rounds can be set to 100~300 rounds. Training is stopped when the validation set loss does not decrease for 10~20 consecutive rounds, and the model parameters corresponding to the lowest validation set loss are saved.

[0094] In actual testing, the channel mean and standard deviation saved during the training phase are used to process the current detection window. Each row of the model output matrix corresponds point by point to the fault equivalent admittance trajectory according to the sampling point number. Then, the probability of the rising phase, the probability of the nonlinear transition phase, and the probability of the falling phase are substituted into the fault initiation form weight formula.

[0095] Finally, the fault initiation morphology weight of the nth sampling point is calculated according to the following formula: In the formula, n is the sampling point number within the detection window. The fault initiation morphology weight for the i-th feeder at the n-th sampling point. , and Let be the probabilities of the rising phase, the nonlinear transition phase, and the falling phase of the i-th feeder at the n-th sampling point, respectively, and μ be the non-negative preset weight coefficient corresponding to the nonlinear transition phase probability. The non-negative preset weight coefficients are the probabilities corresponding to the descent phase.

[0096] Both μ and ν are dimensionless quantities and can take values ​​in the range of 0 to 1. For example, both μ and ν can be set to 0.5, and then adjusted according to the weight distribution of the fault initiation interval and the normal switching descent interval in the validation sample. When μ increases, the enhancing effect of the nonlinear transition phase on the weight of the fault initiation form is improved; when ν increases, the suppressing effect of the descent phase on the weight of the fault initiation form is improved.

[0097] This embodiment constructs the equivalent admittance trajectory of the fault by analyzing the fault residual current signal and the zero-sequence voltage signal, preserving the time-varying relationship between the transient conductance and transient susceptance of each feeder; it uses a time-series segmentation model to determine the change stage corresponding to each sampling point, and combines the probabilities of rising, nonlinear transition and falling stages to generate fault initiation mode weights, highlighting the sampling interval that matches the initiation process of high-resistance grounding faults, while reducing the impact of falling stage data on the subsequent extraction of arc growth slope.

[0098] As an example, the step of extracting frequency band energy dispersion and morphological order consistency variables based on the fault equivalent admittance trajectory, and generating detection results by combining the damped normalized arc growth slope and the model competition discriminant includes: performing time-frequency analysis on the real part of the fault equivalent admittance trajectory to obtain the time-frequency amplitude spectrum; calculating the frequency band energy according to preset frequency bands; normalizing each frequency band energy; and calculating the frequency band energy dispersion based on the normalized frequency band energy; performing nonnegation and smoothing on the real part of the fault equivalent admittance trajectory; calculating the first-order and second-order differences of the smoothed fault equivalent conductance; determining the continuous growth segment, nonlinear transition segment, and continuous decline segment based on the first-order and second-order differences; and determining the continuous growth segment, the nonlinear transition segment, and the... The order of occurrence of the continuously declining segment yields the morphological order consistency variable; the non-negative parts of the damped normalized arc growth slope at each preset time scale are weighted and summed, and the weighted summation result, the model competition discriminant, the frequency band energy dispersion, and the morphological order consistency variable are input into a preset logic function to obtain a comprehensive fault detection score; the maximum value is selected from the damped normalized arc growth slope corresponding to each preset time scale to obtain the maximum value across preset time scales; feeders whose comprehensive fault detection score is higher than a preset score threshold, whose maximum value across preset time scales is higher than a preset slope threshold, whose model competition discriminant is not lower than a preset competition threshold, and whose morphological order consistency variable is a preset consistency value are added to a candidate feeder set, and a detection result is generated based on the candidate feeder set.

[0099] It should be noted that the frequency band energy dispersion refers to a dimensionless quantity calculated based on the normalized energy proportion of the fault equivalent conductance in multiple preset frequency bands; the morphological order consistency variable refers to a binary variable obtained based on the order of occurrence of the continuous growth segment, nonlinear transition segment, and continuous decline segment of the fault equivalent conductance; the comprehensive fault detection score refers to the detection score obtained by combining the damped normalized arc growth slope, model competition discriminant, frequency band energy dispersion, and morphological order consistency variable under each preset time scale; and the candidate feeder set refers to the set of feeder identifiers that meet the conditions of comprehensive fault detection score, maximum value across preset time scales, model competition discriminant, and morphological order consistency.

[0100] It is understandable that, for the i-th feeder, the fault equivalent conductance is selected. As the object of analysis for frequency band energy dispersion, short-time Fourier transform, wavelet transform, or Hilbert-Huang transform are performed on the fault equivalent conductance to obtain the time-frequency amplitude spectrum. The preset analysis frequency range is divided into: One frequency band, of which The set of discrete frequency points corresponding to the l-th frequency band is denoted as . For a power distribution network with a power frequency of 50Hz, the preset analysis frequency range can be selected from 100Hz to 500Hz; when a wider transient frequency range needs to be covered, it can be extended to 100Hz to 1000Hz. The number of frequency bands is determined based on the sampling frequency, the detection window length, and the field waveform data.

[0101] The frequency band energy of the l-th frequency band is: The total bandwidth energy of all preset frequency bands is: In the formula, n is the sampling point number, and h is the frequency discrete point number. Let be the time-frequency amplitude value corresponding to the nth sampling point and the hth frequency discrete point of the i-th feeder. This is the set of sampling point numbers corresponding to the detection window. This is the set of discrete frequency points corresponding to the l-th preset frequency band. Let i be the frequency band energy of the i-th feeder in the l-th frequency band. This represents the total bandwidth energy across all preset frequency bands. When the frequency amplitude spectrum maintains the amplitude dimensions of the fault equivalent conductance, , and preset total frequency band energy threshold All of these have the dimension of Siemens square.

[0102] when When, let the frequency band energy dispersion be 0; when At that time, calculate the normalized energy percentage of the l-th frequency band: Calculate the frequency band energy dispersion based on the normalized energy percentage: In the formula, This represents the normalized energy percentage of the l-th frequency band. The frequency band energy dispersion of the i-th feeder, with a value ranging from 0 to 1; when At that time, the corresponding Process as 0. Determined based on historical normal operation data, historical fault recording data, or field test data.

[0103] The fault equivalent conductance is nonnegated and smoothed: Calculate the first-order and second-order differences: In the formula, Let be the smoothed non-negative fault equivalent conductance of the i-th feeder at the n-th sampling point. For smoothing operators, moving average, median filtering, or low-pass filtering can be used; and These are the first-order and second-order differences of the i-th feeder at the n-th sampling point, respectively, both with Siemens dimensions. If the position of the difference cannot be calculated based on the previous sampling point within the detection window, the corresponding difference value is set to 0.

[0104] When there exists a continuous not less than Each sampling point satisfies When, the starting sampling point of the earliest continuous interval is recorded as When there exists a continuous number of not less than Each sampling point satisfies and When, the starting sampling point of the earliest continuous interval is recorded as When there exists a continuous number of not less than Each sampling point satisfies When, the starting sampling point of the earliest continuous interval is recorded as If no corresponding interval exists, the corresponding sampling point is recorded as positive infinity. Difference threshold. , , and These are the first-order differences for the continuously growing segment, the second-order differences for the nonlinear transition segment, and the differential thresholds for the continuously decreasing segment, respectively. All of these are non-negative numbers with Siemens dimensions, and are determined based on historical normal operation data, historical fault waveform data, or field test data. The minimum number of continuous sampling points is determined based on the sampling frequency and the shortest duration required for the corresponding morphology. , and These represent the minimum number of continuous sampling points corresponding to the continuous growth segment, the nonlinear transition segment, and the continuous decline segment, respectively.

[0105] The sampling points for the initial stage of fault development are: The morphological order consistency variable is: when When this occurs, it indicates that the continuously increasing segment or the nonlinear transition segment appears before the continuously decreasing segment, or that no continuously decreasing segment appears within the detection window; when This indicates that no fault development initiation pattern was detected, or that the continuous decline phase did not occur later than the fault development initiation pattern.

[0106] The overall fault detection score is: In one implementation, the preset logic function is the Sigmoid function: In the formula, The comprehensive fault detection score for the i-th feeder is... Let i be the set of preset time scale indices for which the arc growth slope after effective damping normalization can be obtained for the i-th feeder. The non-negative slope contribution weight for the i-th feeder at the m-th effective preset time scale is given, and the sum of the slope contribution weights corresponding to each effective preset time scale is 1. .

[0107] For bias terms, , , and These are the weighting coefficients for the slope-weighted result, model competition discriminant, frequency band energy dispersion, and morphological order consistency variable, respectively. All input quantities and weighting coefficients are dimensionless, and the weighting coefficients are determined based on training samples labeled with high-resistance grounding faults and normal disturbances.

[0108] when When not empty, calculate the maximum value across a preset time scale; when When empty, the maximum value across the preset time scale is marked as invalid, and the i-th feeder is not added to the candidate feeder set.

[0109] The maximum value across preset time scales is: The candidate feeder set is as follows: In the formula, The set of preset time scale indices for the i-th feeder to obtain the arc growth slope after effective damping normalization; To preset the score threshold, To preset the slope threshold, All three parameters are dimensionless and are used to preset competition thresholds. Each threshold is determined based on the false alarm rate, false negative rate, and engineering tuning requirements in the validation sample.

[0110] When the candidate feeder set is empty, the output shows that no high-resistance grounding fault risk was detected in the current detection window; when the candidate feeder set is not empty, the output shows that there is a high-resistance grounding fault risk in the current detection window and the candidate feeder set, and outputs the comprehensive fault detection score, maximum value across preset time scale, model competition discrimination quantity, frequency band energy dispersion and morphological order consistency variable corresponding to each candidate feeder, without forcing the determination of a unique faulty feeder from the candidate feeder set.

[0111] This example characterizes the energy distribution of fault equivalent conductance across multiple frequency bands through frequency band energy dispersion, determines the morphological change sequence of fault equivalent conductance through first-order and second-order differences, and generates detection results together with arc growth characteristics and model competition discrimination results, reducing misjudgments caused by a single feature exceeding the threshold.

[0112] This application also provides a high-resistance grounding fault detection device based on fault equivalent admittance trajectory. Please refer to [link / reference]. Figure 3 The high-resistance grounding fault detection device based on fault equivalent admittance trajectory includes: The data processing module 10 is used to acquire the zero-sequence voltage and the zero-sequence current at the beginning of each feeder, divide the detection window, and construct the zero-sequence voltage analytical signal and the zero-sequence current analytical signal at the beginning of each feeder. The residual generation module 20 is used to estimate the background zero-sequence admittance and background current analysis signal of each feeder based on the background window, and subtract the background current analysis signal corresponding to each feeder from the zero-sequence current analysis signal at the beginning of each feeder to obtain the fault residual current analysis signal. The trajectory segmentation module 30 is used to construct the fault equivalent admittance trajectory of each feeder based on the fault residual current analytical signal and the zero-sequence voltage analytical signal, and input the fault equivalent admittance trajectory into the time-series segmentation model to obtain the fault initiation morphology weight. The slope processing module 40 is used to generate an enhanced conductance logarithm value from the real part of the fault equivalent admittance trajectory at each preset time scale, extract the arc growth slope according to the fault initiation morphology weight, obtain the resonance damping coefficient, and obtain the damped normalized arc growth slope according to the arc growth slope and the resonance damping coefficient. The model discrimination module 50 is used to obtain the model competition discrimination value based on the fitting results of the arc growth model and the normal shear damping model; The result output module 60 is used to extract the frequency band energy dispersion and morphological order consistency variables based on the fault equivalent admittance trajectory, and generate the detection result by combining the damped normalized arc growth slope and the model competition discriminant.

[0113] The high-resistance grounding fault detection device based on fault equivalent admittance trajectory provided in this application employs the high-resistance grounding fault detection method based on fault equivalent admittance trajectory in the above embodiments. It can solve the technical problem of how to distinguish between high-resistance grounding faults and normal switching transients based on bus zero-sequence voltage and zero-sequence current at the beginning of each feeder. Compared with the prior art, the beneficial effects of the high-resistance grounding fault detection device based on fault equivalent admittance trajectory provided in this application are the same as those of the high-resistance grounding fault detection method based on fault equivalent admittance trajectory provided in the above embodiments. Furthermore, other technical features in the high-resistance grounding fault detection device based on fault equivalent admittance trajectory are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0114] This application provides a high-resistance grounding fault detection device based on fault equivalent admittance trajectory. The high-resistance grounding fault detection device based on fault equivalent admittance trajectory includes: at least one processor; and a memory communicatively connected to at least one processor; wherein the memory stores instructions executable by at least one processor, and the instructions are executed by at least one processor to enable at least one processor to execute the high-resistance grounding fault detection method based on fault equivalent admittance trajectory in the first embodiment described above.

[0115] The following is for reference. Figure 4 This document illustrates a structural schematic diagram of a high-resistance grounding fault detection device based on fault equivalent admittance trajectory suitable for implementing embodiments of this application. The high-resistance grounding fault detection device based on fault equivalent admittance trajectory in this application embodiment can be a distribution automation terminal, a feeder monitoring terminal, a fault recording device, a relay protection and control device, an industrial control computer, or a distribution master station server. The device can be deployed in substations, switching stations, or distribution master stations, and acquires bus zero-sequence voltage, zero-sequence current at the beginning of each feeder, feeder identification, and sampling time through a sampling interface or communication interface. Figure 4The high-resistance grounding fault detection device based on fault equivalent admittance trajectory shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0116] like Figure 4 As shown, a high-resistance ground fault detection device based on fault equivalent admittance trajectories may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in ROM 1002 (Read Only Memory) or a program loaded from storage device 1003 into RAM 1004 (Random Access Memory). RAM 1004 also stores various programs and data required for the operation of the high-resistance ground fault detection device based on fault equivalent admittance trajectories. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, analog sampling interfaces, digital sampling interfaces, or communication interfaces; output devices 1008 including, for example, display interfaces, alarm output interfaces, or communication interfaces; storage devices 1003 including, for example, magnetic tape, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the high-resistance ground fault detection device based on fault equivalent admittance trajectory to communicate wirelessly or wiredly with other devices to exchange data. While the figures show high-resistance ground fault detection devices based on fault equivalent admittance trajectory with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0117] The high-resistance grounding fault detection device based on fault equivalent admittance trajectory provided in this application, employing the high-resistance grounding fault detection method based on fault equivalent admittance trajectory in the above embodiments, can solve the technical problem of how to distinguish between high-resistance grounding faults and normal switching transients based on bus zero-sequence voltage and zero-sequence current at the beginning of each feeder. Compared with the prior art, the beneficial effects of the high-resistance grounding fault detection device based on fault equivalent admittance trajectory provided in this application are the same as those of the high-resistance grounding fault detection method based on fault equivalent admittance trajectory provided in the above embodiments, and other technical features in this high-resistance grounding fault detection device based on fault equivalent admittance trajectory are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0118] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the high-resistivity grounding fault detection method based on fault equivalent admittance trajectory in the above embodiments.

[0119] The computer-readable storage medium provided in this application can be a tangible medium capable of storing computer programs, such as a hard disk, random access memory, read-only memory, flash memory, optical storage device, or magnetic storage device. When the computer program is executed by a processor, it implements the above-described high-resistance grounding fault detection method based on fault equivalent admittance trajectories.

[0120] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by a high-resistance grounding fault detection device based on fault equivalent admittance trajectories, the high-resistance grounding fault detection device based on fault equivalent admittance trajectories causes the device to: acquire zero-sequence voltage and zero-sequence current at the beginning of each feeder; divide the detection window and construct zero-sequence voltage analytical signal and zero-sequence current analytical signal at the beginning of each feeder; estimate the background zero-sequence admittance and background current analytical signal of each feeder based on the background window; subtract the background current analytical signal corresponding to each feeder from the zero-sequence current analytical signal at the beginning of each feeder to obtain the fault residual current analytical signal; and construct the fault residual current analytical signal based on the fault residual current analytical signal and the zero-sequence voltage analytical signal. The fault equivalent admittance trajectory of each feeder is generated and input into a time-series segmentation model to obtain fault initiation morphology weights. At each preset time scale, enhanced conductance logarithmic values ​​are generated from the real part of the fault equivalent admittance trajectory. The arc growth slope is extracted according to the fault initiation morphology weights to obtain the resonant damping coefficient. The damped normalized arc growth slope is obtained based on the arc growth slope and the resonant damping coefficient. The model competition discriminant is obtained based on the fitting results of the arc growth model and the normal switching damping model. The frequency band energy dispersion and morphological sequence consistency variables are extracted from the fault equivalent admittance trajectory. The detection results are generated by combining the damped normalized arc growth slope and the model competition discriminant.

[0121] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described high-resistance grounding fault detection method based on fault equivalent admittance trajectory. This solves the technical problem of distinguishing between high-resistance grounding faults and normal switching transients based on bus zero-sequence voltage and zero-sequence current at the beginning of each feeder. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the high-resistance grounding fault detection method based on fault equivalent admittance trajectory provided in the above embodiments, and will not be repeated here.

[0122] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A high-resistivity grounding fault detection method based on fault equivalent admittance trajectory, characterized in that, The method includes: Obtain the zero-sequence voltage and the zero-sequence current at the beginning of each feeder, divide the detection window, and construct the zero-sequence voltage analytical signal and the zero-sequence current analytical signal at the beginning of each feeder; Based on the background window, estimate the background zero-sequence admittance and background current analytical signal of each feeder, and subtract the background current analytical signal corresponding to each feeder from the zero-sequence current analytical signal at the beginning of each feeder to obtain the fault residual current analytical signal. Based on the fault residual current analytical signal and the zero-sequence voltage analytical signal, the fault equivalent admittance trajectory of each feeder is constructed, and the fault equivalent admittance trajectory is input into the time-series segmentation model to obtain the fault initiation morphology weight. At each preset time scale, an enhanced conductance logarithmic value is generated from the real part of the fault equivalent admittance trajectory. The arc growth slope is extracted according to the fault initiation morphology weight, the resonance damping coefficient is obtained, and the damped normalized arc growth slope is obtained according to the arc growth slope and the resonance damping coefficient. The model competition discriminant is obtained based on the fitting results of the arc growth model and the normal shear damping model; Based on the fault equivalent admittance trajectory, frequency band energy dispersion and morphological order consistency variables are extracted, and detection results are generated by combining the damped normalized arc growth slope and the model competition discriminant.

2. The method as described in claim 1, characterized in that, The steps of acquiring the zero-sequence voltage and the zero-sequence current at the beginning of each feeder, dividing the detection window, and constructing the zero-sequence voltage analytical signal and the zero-sequence current analytical signal at the beginning of each feeder include: The zero-sequence voltage and the zero-sequence current at the beginning of each feeder are sampled synchronously to obtain a discrete sequence of the zero-sequence voltage and a discrete sequence of the zero-sequence current at the beginning of each feeder. The detection window is obtained by dividing the zero-sequence voltage discrete sequence and the zero-sequence current discrete sequence at the beginning of each feeder according to the preset window length and preset sliding step size. Perform a Hilbert transform on the discrete sequence of zero-sequence voltages located within the detection window to obtain the analytical signal of zero-sequence voltages. A Hilbert transform is performed on the discrete sequence of zero-sequence current at the beginning of each feeder located within the detection window to obtain the analytical signal of zero-sequence current at the beginning of each feeder.

3. The method as described in claim 1, characterized in that, The step of estimating the background zero-sequence admittance and background current analytical signal of each feeder based on the background window, and subtracting the background current analytical signal corresponding to each feeder from the zero-sequence current analytical signal at the beginning of each feeder to obtain the fault residual current analytical signal includes: A background window is obtained by selecting consecutive sampling points before the detection window. For each feeder, candidate background zero-sequence admittance is calculated based on the zero-sequence current analytical signal and the zero-sequence voltage analytical signal at the beginning of each feeder. The background zero-sequence admittance of each feeder is updated or frozen based on the average energy of the zero-sequence voltage analysis signal within the background window, the change of the candidate background zero-sequence admittance relative to the previous effective background zero-sequence admittance, and the detection result of the previous detection window containing the background window. The background current analytical signal is obtained based on the background zero-sequence admittance and the zero-sequence voltage analytical signal. The background current analytical signal corresponding to each feeder is subtracted from the zero-sequence current analytical signal at the beginning of each feeder to obtain the fault residual current analytical signal. The formula for calculating the zero-order admittance of the candidate background is as follows: In the formula, i is the feeder number and n is the sampling point number. The set of sampling point numbers corresponding to the background window. The candidate background zero-order admittance for the i-th feed line. Let be the value of the zero-sequence current analytical signal at the beginning of the i-th feeder at the n-th sampling point. The value of the zero-sequence voltage analytical signal at the nth sampling point. This is a complex conjugate operation. The preset voltage squared stability term is positive.

4. The method as described in claim 1, characterized in that, The steps of constructing the fault equivalent admittance trajectory of each feeder based on the fault residual current analytical signal and the zero-sequence voltage analytical signal, and inputting the fault equivalent admittance trajectory into the time-series segmentation model to obtain the fault initiation morphology weights include: For each feeder, the fault equivalent admittance trajectory is constructed based on the fault residual current analytical signal and the zero-sequence voltage analytical signal; Extract the real and imaginary parts of the fault equivalent admittance trajectory to obtain the fault equivalent conductance sequence and the fault equivalent admittance sequence; The fault equivalent conductance sequence and the fault equivalent susceptance sequence are input into the time series segmentation model to obtain the rising phase probability, the nonlinear transition phase probability, and the falling phase probability. The fault initiation mode weight is calculated based on the rising phase probability, the nonlinear transition phase probability, and the falling phase probability, using the following formula: In the formula, i is the feeder number, and n is the sampling point number within the detection window. The fault initiation morphology weight for the i-th feeder at the n-th sampling point. , and These represent the probabilities of the rising phase, the nonlinear transition phase, and the falling phase for the i-th feeder at the n-th sampling point, respectively. These are the non-negative preset weighting coefficients corresponding to the probabilities of the nonlinear transition phase. The non-negative preset weight coefficients are the probabilities corresponding to the descent phase.

5. The method as described in claim 1, characterized in that, The step of extracting frequency band energy dispersion and morphological order consistency variables based on the fault equivalent admittance trajectory, and generating detection results by combining the damped normalized arc growth slope and the model competitive discriminant quantity includes: Time-frequency analysis is performed on the real part of the fault equivalent admittance trajectory to obtain the time-frequency amplitude spectrum. The frequency band energy is calculated according to the preset frequency bands. The frequency band energy is normalized and the frequency band energy dispersion is calculated based on the normalized frequency band energy. The real part of the fault equivalent admittance trajectory is nonnegated and smoothed. The first-order difference and second-order difference of the smoothed fault equivalent conductance are calculated. The continuous growth segment, nonlinear transition segment and continuous decline segment are determined according to the first-order difference and the second-order difference. The morphological order consistency variable is obtained according to the order of appearance of the continuous growth segment, the nonlinear transition segment and the continuous decline segment. The non-negative parts of the damped normalized arc growth slope under each preset time scale are weighted and summed. The weighted summation result, the model competition discrimination quantity, the frequency band energy dispersion, and the morphological order consistency variable are input into a preset logic function to obtain a comprehensive fault detection score. The maximum value is selected from the damped normalized arc growth slope corresponding to each preset time scale to obtain the maximum value across the preset time scale. Feed lines whose comprehensive fault detection score is higher than a preset score threshold, whose maximum value across a preset time scale is higher than a preset slope threshold, whose model competition discrimination value is not lower than a preset competition threshold, and whose morphological order consistency variable is a preset consistency value are added to the candidate feed line set, and detection results are generated based on the candidate feed line set.

6. The method as described in claim 1, characterized in that, The steps of generating enhanced conductance logarithmic values ​​from the real part of the fault equivalent admittance trajectory at each preset time scale, extracting the arc growth slope according to the fault initiation morphology weight, obtaining the resonant damping coefficient, and obtaining the damped normalized arc growth slope based on the arc growth slope and the resonant damping coefficient include: At each of the preset time scales, the peak value of the fault equivalent conductance is extracted from the real part of the fault equivalent admittance trajectory, and the logarithmic value of the enhanced conductance is calculated based on the peak value of the fault equivalent conductance, using the following formula: In the formula, i is the feeder number, m is the preset time scale number, and k is the fault equivalent conductance peak number. Let $T$ be the peak value of the equivalent conductance of the $k$-th fault of the $i$-th feeder at the $m$-th preset time scale. for The non-negative part, This represents the positive increase in the peak equivalent conductance of the k-th fault relative to the peak equivalent conductance of the (k-1)-th fault at the m-th preset time scale for the i-th feeder. The preset difference enhancement coefficient is non-negative. As a positive reference conductance, The pre-defined logarithmic stable term is positive. This is the logarithmic value of the enhanced conductance corresponding to the peak value of the equivalent conductance of the k-th fault in the m-th preset time scale of the i-th feeder. The arc growth slope is obtained by weighting the logarithm of the enhanced conductance with the sampling time according to the weight of the fault initiation mode. The normalized logarithm of the zero-sequence voltage envelope is obtained by dividing the envelope of the zero-sequence voltage analytical signal by the reference voltage. The normalized logarithm of the zero-sequence voltage envelope is inversely proportional to the fitting slope of the sampling time, and compared with zero to obtain the larger value, or the resonant damping coefficient is obtained by calculating based on the system's parallel equivalent zero-sequence damping resistance and the system's equivalent zero-sequence capacitance. The damped, normalized arc growth slope is calculated based on the arc growth slope and the resonant damping coefficient, using the following formula: In the formula, Let be the damped, normalized arc growth slope of the i-th feeder at the m-th preset time scale. Let be the arc growth slope of the i-th feeder at the m-th preset time scale. The preset damping compensation coefficient is non-negative. This is the resonant damping coefficient. The pre-defined normalized stable term is positive.

7. The method as described in claim 6, characterized in that, The step of obtaining the model competition discriminant based on the fitting results of the arc growth model and the normal shear damping model includes: Taking the sampling point where the maximum value of the fault initiation form weight first appears in the detection window as the starting point, the sampling points corresponding to the fault equivalent conductance peak value after the starting point are taken as subsequent peak points, so as to obtain the relative time of the subsequent peak points relative to the starting point and the number of subsequent peak points. The output values ​​of the arc growth model and the normal shear damping model are calculated based on the relative time and the parameters of the model to be fitted, using the following formulas: In the formula, This is the parameter set for the arc growth model. This is the parameter set for the normal shear damping model. The time difference between any of the subsequent peak points and the starting point. The output value for the arc growth model. This is the output value of the normal shear damping model. , , , , , , and These are the parameters of the model to be fitted; When the number of subsequent peak points meets the preset fitting conditions, the arc growth model and the normal switching damping model are fitted according to the fault initiation mode weights to obtain two types of model errors, and the model competition discrimination quantity is obtained according to the two types of model errors and the corresponding number of model parameters.

8. A high-resistivity grounding fault detection device based on fault equivalent admittance trajectory, characterized in that, The device employs the high-resistivity grounding fault detection method based on fault equivalent admittance trajectory as described in any one of claims 1 to 7, and the device comprises: The data processing module is used to acquire zero-sequence voltage and zero-sequence current at the beginning of each feeder, divide the detection window, and construct the zero-sequence voltage analytical signal and the zero-sequence current analytical signal at the beginning of each feeder. The residual generation module is used to estimate the background zero-sequence admittance and background current analytical signal of each feeder based on the background window, and to subtract the background current analytical signal corresponding to each feeder from the zero-sequence current analytical signal at the beginning of each feeder to obtain the fault residual current analytical signal. The trajectory segmentation module is used to construct the fault equivalent admittance trajectory of each feeder based on the fault residual current analytical signal and the zero-sequence voltage analytical signal, and input the fault equivalent admittance trajectory into the time-series segmentation model to obtain the fault initiation morphology weight. The slope processing module is used to generate an enhanced conductance logarithmic value from the real part of the fault equivalent admittance trajectory at each preset time scale, extract the arc growth slope according to the fault initiation morphology weight, obtain the resonance damping coefficient, and obtain the damped normalized arc growth slope based on the arc growth slope and the resonance damping coefficient. The model discrimination module is used to obtain the model competition discrimination value based on the fitting results of the arc growth model and the normal shear damping model; The result output module is used to extract frequency band energy dispersion and morphological order consistency variables based on the fault equivalent admittance trajectory, and generate detection results by combining the damped normalized arc growth slope and the model competition discriminant.

9. A high-resistance grounding fault detection device based on fault equivalent admittance trajectory, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the high-resistivity grounding fault detection method based on fault equivalent admittance trajectories as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the high-resistivity grounding fault detection method based on fault equivalent admittance trajectory as described in any one of claims 1 to 7.

Images