Power distribution network grounding fault positioning method and device, and storage medium

Abstract

The invention discloses a power distribution network ground fault positioning method and device and a storage medium. According to the power distribution network grounding fault positioning method provided by the invention, the grounding fault positioning precision and reliability are remarkably improved by constructing the collaborative optimization signal processing link. Firstly, a self-adaptive denoising algorithm for dynamically adjusting filtering parameters based on error signals is adopted, noise interference in the complex electromagnetic environment of the power distribution network can be effectively restrained, meanwhile, fault features are reserved, and a high-quality signal basis is provided for follow-up processing; secondly, high-frequency transient characteristics are accurately separated through a characteristic extraction filter, instantaneous fluctuation characteristics caused by a ground fault are captured, and the defect that the sensitivity to weak fault characteristics is insufficient in the prior art is overcome; finally, based on an intelligent judgment mechanism of high-frequency transient characteristics, a fault line can be quickly and accurately recognized, and the missed judgment and misjudgment rate is remarkably reduced.

Description

Technical Field

[0001] This invention relates to the technical field of distribution network fault identification, and specifically to a method, device, and storage medium for locating grounding faults in distribution networks. Background Technology

[0002] Fault location in a distribution network can be described as the process of detecting grounding faults caused by short circuits between conductors and the ground in the distribution network using relevant means (e.g., measuring equipment such as current transformers and voltage sensors), and determining the specific location of the fault. This method mainly relies on manual inspection, fixed threshold comparison, or simple current and voltage analysis.

[0003] In related technologies, there is a technical problem of high rates of missed or false fault identification in the scenario of locating grounding faults in distribution networks. Summary of the Invention

[0004] The technical problem this invention aims to solve is the high rate of missed or false detections in fault identification during the location of grounding faults in power distribution networks. The objective is to provide a method, device, and storage medium for locating grounding faults in power distribution networks, thereby resolving the issue of high missed or false detection rates.

[0005] This invention is achieved through the following technical solution:

[0006] In a first aspect, the present invention provides a method for locating grounding faults in a distribution network, comprising:

[0007] Acquire zero-sequence current signals from multiple monitoring points in the power distribution network;

[0008] The zero-sequence current signal is denoised based on a preset adaptive denoising algorithm to obtain a denoised zero-sequence current signal; wherein, the adaptive denoising algorithm is configured to dynamically adjust the filtering parameters based on the error signal to suppress noise;

[0009] The denoised zero-sequence current signal is processed using a preset feature extraction filter to extract high-frequency transient features; wherein, the feature extraction filter is configured to attenuate low-frequency components and retain high-frequency signals above a preset cutoff frequency; wherein, the high-frequency transient features are the features characterizing the instantaneous fluctuation characteristics of ground faults in the high-frequency signals retained by the feature extraction filter.

[0010] The faulty line is determined based on the aforementioned high-frequency transient characteristics.

[0011] Furthermore, the step of acquiring zero-sequence current signals from multiple monitoring points in the distribution network includes:

[0012] The three-phase current at each monitoring point is collected using a pre-set current transformer;

[0013] The corresponding zero-sequence current signal is calculated based on the three-phase current.

[0014] Further, the step of denoising the zero-sequence current signal based on a preset adaptive denoising algorithm to obtain a denoised zero-sequence current signal includes:

[0015] Use the zero-sequence current signal as the input signal;

[0016] Extract the signal sequence within the receptive domain from the input signal;

[0017] The signal sequence within the receptive field is weighted and calculated to obtain the estimated output signal;

[0018] Based on the estimated output signal and the preset expected signal, the error signal is determined;

[0019] Based on the error signal, and combined with the sliding window mechanism, the weights corresponding to the signal sequence within the receptive field are updated.

[0020] The above steps are executed iteratively until the error signal meets the convergence condition, thereby completing the denoising process of the zero-sequence current signal.

[0021] Further, the step of extracting the signal sequence within the receptive domain from the input signal includes:

[0022] A continuous signal sequence is extracted from the input signal through a preset sliding window to form the receptive field region;

[0023] The step of weighting the signal sequence within the receptive domain to obtain the estimated output signal includes:

[0024] Based on a preset least mean square adaptive filtering algorithm, the signal sequence within the receptive field is weighted and calculated to obtain the estimated output signal;

[0025] The step of updating the weights corresponding to the signal sequence within the receptive field region based on the error signal and using a sliding window mechanism includes:

[0026] The updated weight vector is obtained by adding the weight adjustment amount to the original weight vector; wherein the weight adjustment amount is the product of a preset step size factor, the error signal, and the signal sequence within the receptive field.

[0027] Furthermore, the step of processing the denoised zero-sequence current signal using a preset feature extraction filter to extract high-frequency transient features includes:

[0028] The denoised zero-sequence current signal is filtered using a preset elliptical high-pass filter to attenuate the current components at or below the fundamental power frequency and output a high-frequency signal higher than the preset cutoff frequency.

[0029] The transient fluctuation characteristics characterizing the ground fault are extracted from the high-frequency signal as the high-frequency transient features; wherein the high-frequency transient features include at least one of the following: peak value, rising edge, pulse width, abrupt change point, frequency component, oscillation, duration, and amplitude of the zero-sequence current.

[0030] Furthermore, the step of determining the faulty line based on the high-frequency transient characteristics includes:

[0031] Based on the high-frequency transient characteristics of multiple monitoring points, the feature correlation degree between different monitoring points is calculated;

[0032] Based on the aforementioned feature correlation, a fault indication quantity is generated to indicate the fault probability of each monitoring point;

[0033] Based on the fault indication, the faulty line is determined from the plurality of monitoring points.

[0034] Furthermore, the step of calculating the feature correlation degree between different monitoring points based on the high-frequency transient characteristics of multiple monitoring points includes:

[0035] The high-frequency transient characteristics of each monitoring point at each time point are normalized.

[0036] For each pair of monitoring points, the grey relational coefficient of the pair of monitoring points at multiple time points is calculated based on the normalized high-frequency transient feature sequence.

[0037] Based on the calculated grey relational coefficients at each time point, the average grey relational degree between the monitoring points is determined as the feature relational degree.

[0038] Further, the step of generating a fault indication quantity for indicating the fault probability of each monitoring point based on the feature correlation degree includes:

[0039] For each monitoring point, calculate the feature correlation degree between that monitoring point and all other monitoring points, and determine the average feature correlation degree of that monitoring point;

[0040] The average feature correlation degree of each monitoring point is compared with the preset identification threshold, and the comparison result is Booleanized to obtain a Boolean value;

[0041] Based on the Boolean quantity, the cumulative difference value of each monitoring point is calculated as the fault indication quantity; wherein, the cumulative difference value is obtained by summing the absolute value of the difference between the average feature correlation degree of the monitoring point and the average feature correlation degree of each other monitoring point.

[0042] In a second aspect, the present invention provides an electronic device, comprising: a memory, and one or more processors communicatively connected to the memory; the memory stores instructions executable by the one or more processors, the instructions being executed by the one or more processors to cause the one or more processors to implement the method described above.

[0043] Thirdly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0044] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0045] The ground fault location method for power distribution networks provided by this invention significantly improves the accuracy and reliability of ground fault location by constructing a collaboratively optimized signal processing link. First, an adaptive denoising algorithm based on dynamically adjusting filter parameters according to error signals effectively suppresses noise interference in the complex electromagnetic environment of the power distribution network while preserving fault characteristics, providing a high-quality signal foundation for subsequent processing. Second, a feature extraction filter accurately separates high-frequency transient features, capturing the instantaneous fluctuation characteristics caused by ground faults, overcoming the shortcomings of related technologies in terms of insufficient sensitivity to weak fault characteristics. Finally, an intelligent judgment mechanism based on high-frequency transient features can quickly and accurately identify faulty lines, significantly reducing the rate of missed and false detections, thereby shortening fault investigation time and enhancing the operational stability and self-healing capability of the power distribution network under different grounding methods. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0047] Figure 1 This is an architecture diagram of a power distribution network grounding fault location method provided in this embodiment;

[0048] Figure 2 This is one of the flowcharts for a method for locating grounding faults in a power distribution network provided in this embodiment;

[0049] Figure 3 This is the second flowchart of a method for locating grounding faults in a power distribution network provided in this embodiment;

[0050] Figure 4 This is a block diagram of an electronic device provided in this embodiment. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0052] In related technologies, the distribution network is a crucial link in the power system that directly distributes electricity to end users, and its operational reliability is paramount. Single-phase grounding faults are among the most common fault types in distribution networks. To quickly isolate faulty sections and restore power supply, achieving accurate and automatic location of grounding faults has become a core requirement in distribution network operation and maintenance.

[0053] In related technologies, automated fault location often relies on monitoring terminals installed at critical nodes (e.g., feeder outgoing lines, branch points). These terminals acquire three-phase current signals of the power line in real time through current transformers and calculate the zero-sequence current based on these signals. Under normal system operation, the zero-sequence current is theoretically zero; however, when a ground fault occurs, the fault current forms a path through the grounding point, causing a significant change in the zero-sequence current. Therefore, the zero-sequence current is considered a key electrical quantity for ground fault diagnosis and location.

[0054] In related technologies, the core logic of automated positioning is to determine the section where the fault occurred by analyzing the differences or characteristics of the zero-sequence current signals at different monitoring points.

[0055] However, in real-world operating environments, the idealized positioning process described above faces significant challenges. The electromagnetic environment of power distribution networks is complex, and zero-sequence current signals are highly susceptible to noise interference from various sources, including load fluctuations, switching operations, and background harmonics. More critically, ground faults, especially those occurring through transition resistors, generate high-frequency transient currents that, while containing rich fault information, are often weak in amplitude and extremely short in duration. This presents the positioning system with two contradictory requirements: on the one hand, extremely high sensitivity is needed to capture these weak high-frequency transient characteristics; on the other hand, extremely strong anti-interference capabilities are required to suppress noise and prevent false triggering.

[0056] When dealing with this contradiction, related technologies often suffer from the following deficiencies:

[0057] While using ordinary fixed-parameter filters for noise reduction can suppress some noise, their fixed frequency band and parameters cannot adapt to the complex and ever-changing noise environment. Under strong noise, in order to ensure the noise reduction effect, it is often necessary to relax the filtering conditions, which can easily lead to the filtering out of weak high-frequency fault features that are close to the noise frequency band, thus causing missed detections.

[0058] In the feature extraction stage, if a common high-pass filter is used, its transition band is usually relatively flat and its cutoff characteristic is not steep enough. This makes it unable to effectively distinguish fault features from noise in adjacent frequency bands when separating high-frequency transient components, resulting in low signal-to-noise ratio and poor reliability of the extracted features, which can easily lead to misjudgments in subsequent analysis.

[0059] In summary, the relevant technologies have a high rate of missed or false fault identification in the scenario of locating grounding faults in distribution networks.

[0060] like Figure 1 , Figure 2 and Figure 3 As shown, this embodiment provides a method for locating grounding faults in a distribution network. The execution subject of the method can be a smart terminal configured on-site, such as a smart terminal unit, a feeder terminal unit, a fault indicator, or an edge computing device. The smart terminal can be deployed at key monitoring points in the distribution network and directly connected to the current transformer to achieve localized data processing.

[0061] The method can also be implemented by a remote server, which can be deployed in a data center, dispatch center, or cloud platform, etc. It can be an industrial server, virtual machine, or cloud computing instance. This server can receive data transmitted from on-site terminals via a communication network and perform centralized processing and analysis.

[0062] The method may include:

[0063] Step S12: Obtain the zero-sequence current signal from multiple monitoring points in the distribution network.

[0064] In this embodiment, the executing entity can directly or indirectly acquire zero-sequence current signals from multiple monitoring points in the distribution network through a preset sensing device. This sensing device can be a current transformer, a zero-sequence current sensor, or a Rogowski coil, etc. The sensing device can transmit the acquired data to the executing entity via a preset wired or wireless communication network, such as IEC 61850, Modbus RTU / TCP, or DNP3, etc. Data transmission can be periodic or event-triggered, etc.

[0065] Specifically, a zero-sequence current transformer can be pre-installed at each selected monitoring point. The zero-sequence current transformer can be a toroidal iron core wrapped around a three-phase cable, with the three-phase conductors being measured (forming the primary winding) on ​​its primary side and a measuring winding wound around the iron core on its secondary side. According to the principle of electromagnetic induction, when the three-phase currents are unbalanced, the resulting zero-sequence magnetic flux will form a path in the iron core, thereby inducing a current signal proportional to the zero-sequence current on the secondary side.

[0066] Alternatively, three-phase current transformers can be pre-installed at each selected monitoring point to independently acquire the currents of phases A, B, and C. Each phase current passes through its own independent signal conditioning circuit (amplification and anti-aliasing filtering) and analog-to-digital converter to obtain three discrete digital sequences. The preset algorithm in the execution unit can perform predetermined mathematical operations on these three sequences to obtain a new digital sequence (zero-sequence current signal).

[0067] In this embodiment, the monitoring point can be represented as a spatial location in the distribution network topology selected for installing the data acquisition device. Specifically, it can be the feeder outlet of a substation, the sectionalizing switch of a main line, the user access point, or the inlet of a distribution transformer, etc.

[0068] In this embodiment, the zero-sequence current signal can be represented as a composite current component obtained through vector operations in a three-phase AC power system due to the asymmetric components present in the A, B, and C phase currents. It is understood that in the scenario of a single-phase ground fault in a distribution network, the faulty phase forms a loop with the ground through the grounding point, severely disrupting the symmetry of the three-phase currents and generating a significant zero-sequence current.

[0069] Step S14: Denoise the zero-sequence current signal based on a preset adaptive denoising algorithm to obtain a denoised zero-sequence current signal; wherein, the adaptive denoising algorithm is configured to dynamically adjust the filtering parameters based on the error signal to suppress noise.

[0070] In this embodiment, the preset adaptive denoising algorithm can be described as a denoising algorithm that can automatically and online adjust its internal core parameters. That is, it makes the output signal of the algorithm continuously approach a desired signal. In this embodiment, the desired signal is an ideal zero-sequence current without noise.

[0071] In one specific implementation, the adaptive denoising algorithm can be an adaptive filter based on the least mean square criterion. An FIR filter structure can be used, with coefficient updates achieved through the LMS algorithm. Specifically, the zero-sequence current signal is first input into the transverse filter to obtain an estimated output signal; then, the estimated output is compared with the desired signal to generate an error signal; finally, the filter tap weights are adjusted based on the magnitude of the error signal.

[0072] In one specific implementation, the adaptive denoising algorithm can be a recursive least squares adaptive algorithm. Specifically, matrix operations can be used to achieve optimal filtering by updating the filter coefficients by minimizing the weighted sum of squared errors.

[0073] In a specific implementation, the adaptive denoising algorithm may be a normalized least mean square algorithm, an affine projection algorithm, a frequency domain adaptive filtering algorithm, or a kernel-based adaptive filtering algorithm, etc.

[0074] In this embodiment, noise suppression can be expressed as reducing the amplitude of non-fault-induced, irrelevant random fluctuations or periodic interferences in the zero-sequence current signal. This noise can originate from electronic noise from sensing devices, electromagnetic coupling between adjacent lines, load switching operations, or high-frequency harmonics generated by power electronic equipment, etc.

[0075] Step S16: Process the denoised zero-sequence current signal using a preset feature extraction filter to extract high-frequency transient features; wherein, the feature extraction filter is configured to attenuate low-frequency components and retain high-frequency signals above a preset cutoff frequency; wherein, the high-frequency transient features are the features characterizing the instantaneous fluctuation characteristics of the ground fault in the high-frequency signal retained by the feature extraction filter.

[0076] In this embodiment, the feature extraction filter can be represented as a signal processing algorithm or hardware unit for achieving frequency selectivity. It can significantly attenuate components of the input signal below a certain frequency (cutoff frequency) while allowing components above the cutoff frequency to pass through as undistortedly as possible.

[0077] In this embodiment, the low-frequency components can be represented as the fundamental power frequency current of the distribution network and its main integer harmonics. Low-frequency components often have large amplitudes in the zero-sequence current, but are mainly caused by normal load imbalances and are not specific characteristics of ground faults. The goal of the feature extraction filter is to suppress low-frequency components to a sufficiently low level to prevent them from overwhelming or interfering with weak fault transient characteristics.

[0078] In this embodiment, the preset cutoff frequency can be a threshold frequency used to ensure that the filter only retains the frequency band relevant to the fault, reducing interference from irrelevant noise. This preset cutoff frequency can be pre-set based on the actual spectral distribution of the high-frequency transient current of the ground fault.

[0079] In this embodiment, the high-frequency signal can be represented as the output signal obtained by the filter performing convolution operation on the input denoised zero-sequence current signal or by sliding window filtering.

[0080] In this embodiment, the high-frequency transient characteristic can be a set of parameters, which can be one or more physical parameters calculated from the high-frequency signal output by the feature extraction filter to quantitatively describe the instantaneous current fluctuation characteristics when a ground fault occurs. Specifically, it can include: peak value, rising edge, pulse width, abrupt change point, frequency components, oscillation, duration, and amplitude, etc.

[0081] In one specific implementation, the feature extraction filter can be an elliptic function filter. Specifically, an elliptic function filter has a steep transition band and balanced passband and stopband ripples. In this embodiment, it can be used for feature extraction.

[0082] In one specific implementation, the feature extraction filter may be a Butterworth filter, a Chebyshev filter, or a finite impulse response filter, etc.

[0083] In one specific implementation, the feature extraction filter can also be an analog filter circuit.

[0084] Step S18: Determine the faulty line based on the high-frequency transient characteristics.

[0085] In this embodiment, the executing entity can identify faulty lines by analyzing the distribution differences of high-frequency transient characteristics among different monitoring points. Specifically, firstly, the correlation or similarity between each pair of high-frequency transient characteristic sequences (e.g., transient energy sequences or dominant frequency sequences) of all monitoring points is calculated. Grey relational analysis can be used. More specifically, the characteristic sequence of each feeder (monitoring point) can be regarded as a grey system factor, and the similarity of their waveform development trends can be quantified by calculating the grey relational degree between the sequences of each factor. The higher the correlation, the more synchronized the transient behavior of the two feeders is, and the more likely they are in the same operating state (normal or faulty). Then, based on the calculated correlation matrix, a quantitative index indicating the probability of fault is generated for each monitoring point (or the feeder it represents), i.e., a fault indication quantity.

[0086] In this embodiment, the executing entity can also directly compare the amplitudes of the same type of high-frequency transient characteristics extracted from different monitoring points. For example, comparing the peak value, total transient energy, or root mean square value of the high-frequency transient current at each monitoring point. The line corresponding to the monitoring point with the largest characteristic amplitude is identified as the faulty line. Alternatively, a minimum amplitude threshold can be set to prevent misjudgment in the absence of faults or minor disturbances.

[0087] In this embodiment, the executing entity can also use intelligent algorithms such as machine learning to establish a mapping model between features and fault locations by training historical fault data, thereby achieving intelligent fault location.

[0088] The ground fault location method for power distribution networks provided by this invention significantly improves the accuracy and reliability of ground fault location by constructing a collaboratively optimized signal processing link. First, an adaptive denoising algorithm based on dynamically adjusting filter parameters according to error signals effectively suppresses noise interference in the complex electromagnetic environment of the power distribution network while preserving fault characteristics, providing a high-quality signal foundation for subsequent processing. Second, a feature extraction filter accurately separates high-frequency transient features, capturing the instantaneous fluctuation characteristics caused by ground faults, overcoming the shortcomings of related technologies in terms of insufficient sensitivity to weak fault characteristics. Finally, an intelligent judgment mechanism based on high-frequency transient features can quickly and accurately identify faulty lines, significantly reducing the rate of missed and false detections, thereby shortening fault investigation time and enhancing the operational stability and self-healing capability of the power distribution network under different grounding methods.

[0089] In some implementations, the step of acquiring zero-sequence current signals from multiple monitoring points in the distribution network includes:

[0090] Step S122: Collect the three-phase current at each monitoring point using a preset current transformer.

[0091] In this embodiment, a current transformer (CT) is a device used to measure current. By converting high-current signals into low-current signals, it enables power system monitoring, protection, and measurement equipment to process current information safely and accurately. The monitoring point is the location in the distribution network where the current transformer is installed, used to collect current data in real time. Three-phase current refers to the three-phase alternating current in the power system, usually labeled A, B, and C, representing three current signals with a phase difference of 120 degrees.

[0092] It should be noted that collecting three-phase current signals from multiple monitoring points in the distribution network using current transformers can accurately reflect the network's operating status in real time. By deploying current transformers at different monitoring points, comprehensive monitoring of all areas of the distribution network can be achieved, ensuring the comprehensiveness and accuracy of the data. This process provides a solid foundation for subsequent zero-sequence current calculations, making the entire fault location process more reliable and accurate. Real-time data acquisition helps to quickly detect any anomalies or faults in the power grid, providing precise information for subsequent processing steps (such as noise reduction and feature extraction), thereby improving the speed and accuracy of fault diagnosis.

[0093] In one possible implementation, the three-phase current specifically includes: phase A current, phase B current, and phase C current.

[0094] Step S124: Calculate the corresponding zero-sequence current signal based on the three-phase current.

[0095] In this embodiment, the zero-sequence current signal is a current component in a three-phase power system, typically caused by asymmetry or imbalance of the three-phase currents. It is the vector sum of the three phase currents (phase A, phase B, and phase C), and often increases significantly during ground faults. Therefore, zero-sequence current is a key characteristic of ground faults.

[0096] It should be noted that calculating the zero-sequence component of the three-phase current provides an important analytical basis for subsequent fault diagnosis. Ground faults usually cause abnormal changes in the zero-sequence current; therefore, accurately calculating the zero-sequence current is the core of fault location.

[0097] In one possible implementation, the formula for calculating the zero-sequence current is as follows:

[0098] ;

[0099] In the formula, Represents zero-sequence current. Indicates the current of phase A. This represents the B-phase current. This represents the C-phase current.

[0100] It should be noted that by analyzing the zero-sequence component of the three-phase current, the unique current characteristics caused by faults in the power grid can be effectively extracted, laying the foundation for subsequent signal processing, feature extraction, and fault location. Furthermore, the calculation of the zero-sequence current does not require additional hardware support, relying solely on existing current measurement data, making this step highly efficient and low-cost, and it can be completed quickly during real-time monitoring.

[0101] In some embodiments, the step of denoising the zero-sequence current signal based on a preset adaptive denoising algorithm to obtain a denoised zero-sequence current signal includes:

[0102] Step S142: Use the zero-sequence current signal as the input signal.

[0103] In this embodiment, the executing entity can use the zero-sequence current signal as the input signal of the variable step size LMS adaptive filter.

[0104] Specifically, the variable step size LMS adaptive filter is an adaptive filter used in signal processing. It optimizes signal processing by adjusting the filter coefficients to minimize the error between the input signal and the desired output. Variable step size refers to the step size factor being dynamically adjusted as the signal characteristics change, thereby achieving better adaptability and denoising performance. Denoising processing refers to the process of removing interference or noise from a signal, with the aim of preserving useful information and making subsequent analysis and processing more accurate.

[0105] It should be noted that using a variable-step-size LMS adaptive filter for denoising effectively improves the quality of the zero-sequence current signal. Zero-sequence current signals in power systems are often subject to noise interference, such as external electromagnetic interference and equipment errors, which can affect the accuracy of fault analysis. By using a variable-step-size LMS adaptive filter, the filter parameters can be adjusted in real time to adapt to different noise levels, thereby extracting the useful signal more accurately. Compared to a fixed-step-size LMS filter, the variable-step-size LMS filter can more flexibly respond to signal changes, improving the filtering effect and the accuracy of signal recovery.

[0106] Step S144: Extract the signal sequence within the receptive field region from the input signal.

[0107] In this embodiment, the executing entity can select a finite-length subsequence from the continuous input signal for processing using a preset algorithm. The receptive field can be a dynamic window covering the signal segment near the current moment, and its size can be determined by the parameters of the preset algorithm. The purpose of extraction is to isolate local signals that may contain fault features or noise, providing input for subsequent weighted calculations.

[0108] Specifically, the receptive field can be a fixed-length sliding window, a variable-length adaptive window, or an event-triggered dynamic window. For example, the sliding window length can be set to 100-500 sampling points to balance processing accuracy and computational overhead.

[0109] In a specific implementation, the receptive domain refers to the selected, important portion of the input signal for further processing. In signal processing, it's common practice to focus on certain features or changes in a signal while ignoring irrelevant components.

[0110] In one specific implementation, a sliding window of a preset length can be used. At each discrete time step, this window covers M consecutive sample values ​​of the input signal from the current time step back to the previous M-1 time steps, thus forming an M-dimensional column vector. Here, M is the order of the filter, i.e., the size of the receptive field, which is a preset design parameter. This M-dimensional column vector is the signal sequence extracted from the input signal within the receptive field.

[0111] Step S146: Perform weighted calculations on the signal sequence within the receptive field region to obtain the estimated output signal.

[0112] In this embodiment, the weighted calculation means performing a vector inner product operation between the receptive field signal vector obtained in step S144 and the current weight vector of the filter.

[0113] In a specific implementation plan, the weighted calculation can be performed using the following formula:

[0114] ;

[0115] In the formula, Let represent the estimated output signal of the filter at time t during the nth iteration. This represents the signal value of the input signal at time t-k+1. This represents the k-th weight in the weight vector at time t. This represents the receptive field region extracted at time t. Let T represent the updated weight vector of the filter at time t during the nth iteration, where T represents the transpose and N represents the total number of weights.

[0116] Step S148: Determine the error signal based on the estimated output signal and the preset expected signal.

[0117] In this embodiment, the executing entity can compare a preset desired signal with the estimated output signal obtained in the previous step. The desired signal can be a fixed value close to zero, that is, it is assumed that the zero-sequence current should be zero when there is no fault or the noise is stable.

[0118] In this embodiment, the error signal can be the difference between the estimated output signal and a preset desired signal. This error signal represents the gap between the current performance and the desired performance of the filter, which includes both unfiltered noise and signal distortion caused by improper filtering.

[0119] In a specific implementation scheme, the error signal can be calculated using the following formula:

[0120] ;

[0121] In the formula, This represents the error signal at time t during the nth iteration. This represents the desired signal at time t.

[0122] The error signal refers to the difference between the desired signal and the actual estimated output. The filter continuously adjusts the weights by minimizing the error, making the output signal closer to the desired signal.

[0123] Step S1410: Based on the error signal, and using a sliding window mechanism, update the weights corresponding to the signal sequence within the receptive field region.

[0124] In this implementation, the executing entity can adjust the weight coefficients using the magnitude and direction of the error signal, while considering the time range of the sliding window. The sliding window mechanism ensures that weight updates are based only on the most recent signal data, maintaining the algorithm's real-time performance and adaptability. The update process can make the estimated output closer to the desired signal, thereby reducing errors.

[0125] In this embodiment, the least mean square algorithm can be used to update the weights corresponding to the signal sequence within the receptive field region. Alternatively, the normalized least mean square algorithm, the recursive least squares algorithm, or the affine projection algorithm can be used to update the weights corresponding to the signal sequence within the receptive field region.

[0126] In this embodiment, the size and step size of the sliding window can be preset, and the windows can overlap or not overlap. For example, the window length can be set to 50 sampling points, and the step size can be 1 sampling point to achieve point-by-point updates. The step size factor can be fixed or adaptively varied, for example, dynamically adjusted based on signal power to balance convergence speed and stability.

[0127] In a specific implementation plan, the weight update can be performed using the following formula:

[0128] ;

[0129] In the formula, This represents the updated weight vector of the filter at time t+1 during the nth iteration. Let μ represent the updated weight vector of the filter at time t during the nth iteration, and μ represent the step size factor.

[0130] Step S1412: Iteratively execute the above steps until the error signal meets the convergence condition to complete the denoising process of the zero-sequence current signal.

[0131] In this embodiment, steps S144 to S1410 can be repeated to achieve an estimation-error-update cycle.

[0132] In this embodiment, the convergence condition is a preset criterion for determining whether the algorithm should stop iterating. Specifically, it can be that the mean square value or average power of the error signal changes less than a preset threshold over time. It can also be that the update amount (Euclidean norm) of the weight vector is less than a certain minimum value, or it can be that a fixed number of iterations are performed.

[0133] It should be noted that the variable step-size LMS adaptive filter can dynamically adjust the step size according to signal changes, thus providing flexible adaptability in different noise environments. By extracting the region of interest, the filter can focus on the key parts of the signal and ignore irrelevant noise, thereby improving processing efficiency and accuracy. Weighted calculation enhances the influence of important signals, helping to improve the filter's sensitivity to ground fault characteristics. Through the sliding window mechanism, the filter can optimize weights in different time periods, thereby responding to changes in the zero-sequence current signal and improving the stability and real-time performance of the filter. Finally, through continuous iteration and error minimization, the filter ensures the denoising effect of the zero-sequence current signal, thus providing clearer and more accurate data support for subsequent fault location steps.

[0134] In some implementations, the step of extracting the signal sequence within the receptive domain from the input signal includes:

[0135] Step S1442: Extract a continuous signal sequence from the input signal through a preset sliding window to form the receptive field region.

[0136] In this embodiment, the sliding window can be a first-in, first-out (FIFO) circular buffer. This buffer occupies a fixed amount of space in memory, storing exactly M sample values ​​(M being the window length). When a new sample point arrives, the oldest sample point is removed from the buffer, and the new point is filled in, thus realizing the sliding of the window. The window length M is a preset value, which can be pre-set based on prior knowledge of the noise and fault transient characteristics in the zero-sequence current signal.

[0137] The step of weighting the signal sequence within the receptive domain to obtain the estimated output signal includes:

[0138] Step S1462: Based on the preset least mean square adaptive filtering algorithm, the signal sequence within the receptive field is weighted and calculated to obtain the estimated output signal.

[0139] In this embodiment, the least mean square adaptive filtering algorithm is an adaptive algorithm with the optimization objective of minimizing the instantaneous mean square error. It adjusts the filter weights to minimize the instantaneous estimate (or short-term average) of the square of the error between the filter output signal and the desired signal.

[0140] The step of updating the weights corresponding to the signal sequence within the receptive field region based on the error signal and using a sliding window mechanism includes:

[0141] Step S14102: Add the weight vector before the update to the weight adjustment amount to obtain the updated weight vector; wherein, the weight adjustment amount is the product of a preset step size factor, the error signal, and the signal sequence within the receptive field.

[0142] In this embodiment, the weight adjustment amount is a vector, the magnitude of which and its direction together determine the magnitude and direction of the weight update. Specifically, it may include:

[0143] The preset step size factor is a preset, positive constant that controls the step size or learning rate of each weight update.

[0144] The error signal provides directional guidance and magnitude information needed for weight updates. The sign of the error (positive or negative) indicates in which direction the weights should be adjusted to reduce the error. The magnitude of the error reflects the degree of inaccuracy of the current estimate. A larger error generally means a larger weight adjustment is required.

[0145] The signal sequence within the receptive field, which is the reference input vector extracted through a sliding window in step S1442, determines how the overall adjustment is distributed to the specific components of the weight vector. Each element in the signal sequence within the receptive field participates in calculating the adjustment of its corresponding weight. That is, at a certain moment, the input signal component with a larger amplitude will exert a greater adjustment influence on its corresponding weight coefficient.

[0146] like Figure 2 As shown, in some embodiments, the step of processing the denoised zero-sequence current signal using a preset feature extraction filter to extract high-frequency transient features includes:

[0147] Step S162: Use a preset elliptical high-pass filter to filter the denoised zero-sequence current signal to attenuate the current components at or below the fundamental power frequency and output a high-frequency signal higher than the preset cutoff frequency.

[0148] In this embodiment, the elliptic high-pass filter is an IIR (Infinite Impulse Response) digital filter with equivalent ripple characteristics in both the passband and stopband. It can achieve the steepest transition band of all filter types for a given filter order. That is, it can achieve a rapid transition from high suppression to almost no attenuation near a preset cutoff frequency with extremely high frequency resolution.

[0149] Step S164: Extract the instantaneous fluctuation characteristics characterizing the ground fault from the high-frequency signal as the high-frequency transient features; wherein, the high-frequency transient features include at least one of the following: peak value, rising edge, pulse width, abrupt change point, frequency component, oscillation, duration, and amplitude of the zero-sequence current.

[0150] In this embodiment, the executing entity can perform calculations and analyses on the high-frequency signal sequence based on a preset signal processing method, thereby outputting one or more parameters that can numerically describe the instantaneous fluctuation characteristics of the signal. For example, an analysis time window can be determined based on a fault initiation detection algorithm (using a threshold method or wavelet transform modulus maxima method), and the instantaneous fluctuation characteristics characterizing the ground fault can be extracted through this analysis time window.

[0151] Specifically, the peak value represents the maximum absolute value of the high-frequency transient signal amplitude within the analysis time window. This maximum value can be found by traversing all sampling points within the analysis window. The peak value directly reflects the maximum current surge intensity caused by a ground fault. A higher peak value indicates a smaller fault point impedance or a larger system-to-ground capacitance.

[0152] The rising edge represents the time it takes for a high-frequency transient signal to rise from a low percentage (e.g., 10%) of its peak value to a high percentage (e.g., 90%). It characterizes the steepness and speed of fault current buildup. Shorter rising edges can correspond to transient faults or sudden impedance changes at the fault point. First, the peak value and its occurrence time can be determined. Then, a forward search is performed to find the moments when the signal first crosses 0.1 times and 0.9 times the peak value; the rising edge is the difference between these two moments.

[0153] Pulse width represents the duration for which the envelope of a high-frequency transient signal remains at a relatively high level (e.g., exceeding 50% of its peak value). The signal envelope can be calculated (using Hilbert transform or moving average) to determine the start and end times when the envelope exceeds 0.5 times the peak value; the time difference between these times is the pulse width.

[0154] Abrupt change points represent moments when a signal undergoes a dramatic change, particularly the onset of a fault. Abrupt change point detection algorithms can be employed, such as wavelet transform modulus maxima detection (using local extrema of wavelet coefficient moduli to locate singularities) or Teager-Kaiser energy operator analysis (which can sharpen the transient characteristics of a signal).

[0155] Frequency components are represented as the dominant oscillation frequencies or their distribution that constitute a high-frequency transient signal. A Fast Fourier Transform can be performed on the transient signal segment within the analysis window to obtain its spectrum, and then the frequency corresponding to the main peak in the spectrum can be extracted; alternatively, a Short-Time Fourier Transform can be used to observe the frequency change over time.

[0156] Oscillations are parameters that describe the damped oscillation characteristics of a transient process, such as the dominant oscillation frequency and the damping factor. A transient signal can be modeled as a damped sine wave, and the oscillation frequency and damping factor can be fitted using a parameter identification algorithm (e.g., the Prony algorithm).

[0157] Duration represents the total time from the point of sudden fault initiation to the decay of the high-frequency transient signal to the level of background noise.

[0158] Amplitude is represented as a measure of the overall intensity of a high-frequency transient signal. It can be used to calculate the root mean square value of the signal within an analysis window, or to calculate the total energy of the signal (by numerically integrating the square of the signal).

[0159] This implementation uses an elliptical high-pass filter with extremely steep cutoff characteristics to filter out strong fundamental power frequency and low-frequency harmonic interference with high precision and selectivity. This effectively separates the weak but crucial high-frequency transient signal containing key fault information. Furthermore, multi-dimensional instantaneous fluctuation characteristics, including peak value, rising edge, and frequency components, are extracted from this pure high-frequency signal. This provides a rich, accurate, and interference-resistant data foundation for fault analysis, significantly improving the detection sensitivity and location reliability of grounding faults, especially high-resistance faults and weak transient faults, in complex noise environments.

[0160] In one specific implementation, multiple high-frequency transient current characteristics that trigger ground faults can be extracted based on the denoised zero-sequence current using an elliptic high-pass filter. An elliptic high-pass filter is a signal processing tool that allows signals with frequencies above a specific cutoff frequency to pass through while suppressing signals with frequencies below that frequency. In ground fault analysis, the high-pass filter can effectively filter out low-frequency normal operating currents, retaining only the high-frequency transient signals caused by the fault. High-frequency transient current characteristics refer to the higher-frequency components of the current signal, typically related to transient phenomena generated when electrical equipment malfunctions (such as ground faults). High-frequency transient signals can reflect short-term current changes in the power grid and have high diagnostic value.

[0161] It should be noted that by using a high-pass filter to extract high-frequency transient current characteristics from the zero-sequence current signal, the transient and rapidly changing current signal caused by a ground fault can be effectively separated. Ground faults are often accompanied by instantaneous high-frequency current fluctuations, while the low-frequency components may be normal operating currents of the power grid. A high-pass filter can filter out these low-frequency signals, allowing for focused analysis of the high-frequency components caused by the fault. This enables the system to more accurately capture fault characteristics, reduce noise interference, and improve signal reliability and diagnostic accuracy.

[0162] In one possible implementation, the high-frequency transient current characteristics specifically include: the peak value of the zero-sequence current, the rising edge, the pulse width, the abrupt change point, the frequency components, the oscillation, the duration, and the amplitude.

[0163] It should be noted that by extracting these high-frequency transient current characteristics, the system can identify fault modes more quickly, providing more accurate data support for subsequent fault analysis and location, thereby greatly improving the detection efficiency and accuracy of grounding faults.

[0164] In some implementations, the step of determining the faulty line based on the high-frequency transient characteristics includes:

[0165] Step S182: Calculate the feature correlation degree between different monitoring points based on the high-frequency transient characteristics of multiple monitoring points.

[0166] In this embodiment, this step reveals the spatial distribution pattern of faults in the distribution network by quantitatively analyzing the similarity or correlation between high-frequency transient characteristics collected from different monitoring points. That is, the transient characteristic behavior of a line experiencing a ground fault will differ significantly from that of a normal line.

[0167] In this embodiment, the feature correlation degree can be represented as a numerical index used to measure the similarity or correlation strength between the high-frequency transient feature sequences of two monitoring points. The higher the value, the more synchronized and similar the transient behaviors of the two monitoring points are; the lower the value, the greater the difference in their behaviors.

[0168] In this embodiment, grey relational analysis can be used to calculate the feature correlation degree between different monitoring points. Specifically, firstly, the high-frequency transient features extracted from each monitoring point (e.g., transient energy sequence, dominant frequency sequence, or feature vector sequence composed of multiple features) are used as the behavioral feature sequence of that monitoring point. Then, to eliminate the influence of different feature dimensions and orders of magnitude, each feature sequence can be normalized, for example, by using mean transformation. Next, for any two monitoring points' feature sequences, the corresponding grey relational coefficient can be calculated at each time step. Finally, the average of the correlation coefficients of the two sequences over all time steps is taken to obtain the feature correlation degree between the two monitoring points.

[0169] Step S184: Based on the feature correlation degree, generate a fault indication quantity to indicate the fault probability of each monitoring point.

[0170] In this embodiment, the fault indication quantity can be a scalar value assigned to each monitoring point, the magnitude of which is positively correlated with the probability of a ground fault occurring on the line where the monitoring point is located. That is, the transient characteristic behavior of a faulty line will be significantly different from that of most normal lines. Specifically, for a certain monitoring point, the average value of its characteristic correlation degree with all other monitoring points in the system can be calculated. This value reflects the behavioral similarity between the corresponding monitoring point and the system as a whole. Under normal circumstances, the correlation degree of each point is high; when a fault occurs, the correlation degree between the faulty point and normal points will decrease, thereby causing its average correlation degree to decrease. To amplify the abnormality of the faulty point, the absolute value of the difference between the average correlation degree of the monitoring point and the average correlation degree of all other monitoring points can also be calculated as its fault indication quantity.

[0171] Step S186: Determine the faulty line from the plurality of monitoring points based on the fault indication quantity.

[0172] In this implementation, the executing entity can iterate through all the fault indication values ​​calculated from all monitoring points and identify the one with the largest value. The line corresponding to this monitoring point is the line with the most abnormal behavior and is therefore the most likely faulty line.

[0173] To improve anti-interference capability and prevent false alarms, the maximum value of the fault indication can be compared with a preset identification threshold. This identification threshold can be preset based on historical data, simulation analysis, or field experience. The identification threshold can also be adaptively adjusted according to the network structure and grounding method (e.g., a higher threshold is used for low-resistance grounding systems, and a lower threshold is used for arc suppression coil grounding systems).

[0174] If the maximum value of the fault indication is greater than or equal to the identification threshold, the line corresponding to the monitoring point is determined to be a faulty line. Conversely, if it is less than the threshold, it can be determined that no clear fault has occurred or the fault is not within the monitoring range, thus avoiding false alarms.

[0175] In some implementations, the step of calculating the feature correlation degree between different monitoring points based on the high-frequency transient characteristics of multiple monitoring points includes:

[0176] Step S1822: Normalize the high-frequency transient characteristics of each monitoring point at each time step.

[0177] In this embodiment, the normalization process means mapping the original high-frequency transient feature sequence to a unified, dimensionless numerical range, so that the feature sequences of different monitoring points are comparable.

[0178] In a specific implementation plan, the high-frequency transient characteristics of each monitoring point at each time step can be normalized using the following formula:

[0179] ;

[0180] In the formula, This represents the normalized zero-sequence current transient characteristic of the i-th feeder at time k. Let i represent the transient characteristics of the original zero-sequence current of the i-th feeder at time k, where i = 1, 2, ..., l, and l represents the total number of feeders.

[0181] Step S1824: For each pair of monitoring points, calculate the grey relational coefficient of the pair of monitoring points at multiple times based on the normalized high-frequency transient feature sequence.

[0182] In this embodiment, the grey relational coefficient is represented as a scalar between 0 and 1, used to measure the closeness between the normalized feature values ​​of two detection points at a certain moment. The closer the coefficient value is to 1, the more synchronized and similar the behavior of the two sequences at that moment.

[0183] In a specific implementation plan, the grey relational coefficient of the pair of monitoring points at multiple times can be calculated using the following formula:

[0184] ;

[0185] ;

[0186] ;

[0187] In the formula, Let represent the grey relational coefficient between the i-th feeder and the j-th feeder at time k, where k = 1, 2, ..., n, and n represents the total number of times. This represents the minimum change in the zero-sequence current transient characteristic of all monitoring points across different monitoring points. The value represents the maximum variation of the zero-sequence current transient characteristics at all monitoring points across different monitoring points, where ρ represents the resolution coefficient. This represents the normalized zero-sequence current transient characteristic of the j-th feeder at time k. This represents the normalized zero-sequence current transient characteristic of the i-th feeder at time k. This represents the difference in the zero-sequence current transient characteristics between the i-th and j-th feeders at time k. This represents the minimum difference in the transient characteristics of the zero-sequence current between different feeders. This indicates that the minimum value is taken over all time intervals.

[0188] Step S1826: Based on the calculated grey relational coefficients at each time point, determine the average grey relational degree between the pair of monitoring points, which is used as the feature relational degree.

[0189] In this embodiment, this step can aggregate the time-varying local correlation coefficient sequence obtained in step S1824 into a single indicator that can comprehensively reflect the overall correlation strength between the two monitoring points throughout the entire analysis period.

[0190] In this embodiment, the feature correlation degree is a quantitative representation of the global similarity between any two monitoring points. It is obtained by averaging the local correlation coefficients at each time point, reflecting the degree of similarity in the overall development trend and shape of the high-frequency transient feature sequences of the two monitoring points.

[0191] In a specific implementation scheme, the average gray correlation degree between the pair of monitoring points can be determined by the following formula, which serves as the feature correlation degree:

[0192] ;

[0193] In the formula, This represents the similarity of the zero-sequence current transient characteristics between the i-th feeder (i.e., the i-th monitoring point) and the j-th feeder (i.e., the j-th monitoring point).

[0194] In some implementations, the step of generating a fault indication quantity for indicating the fault probability of each monitoring point based on the feature correlation degree includes:

[0195] Step S1842: For each monitoring point, calculate the feature correlation degree between the monitoring point and other monitoring points, and determine the average feature correlation degree of the monitoring point.

[0196] In this embodiment, the executing entity can iterate through the feature correlation degrees between each monitoring point in the system and all other monitoring points in the system. The arithmetic mean of these correlation degrees can be calculated as the average feature correlation degree of that monitoring point.

[0197] In a specific implementation plan, the characteristic correlation degree between the monitoring point and other monitoring points can be calculated using the following formula:

[0198] ;

[0199] In the formula, This represents the average similarity between the nth feeder and all other feeders. The similarity between the nth feeder and the jth feeder is represented by L, where L represents the total number of feeders. This represents the average similarity.

[0200] Step S1844: Compare the average feature correlation degree of each monitoring point with the preset identification threshold, and perform Boolean processing on the comparison result to obtain a Boolean quantity.

[0201] In this embodiment, the preset identification threshold can be a threshold value pre-set based on historical operating data, simulation analysis, or field experience. Its function is to define at what level the average correlation of a monitoring point is considered to have sufficiently abnormal behavior and deserves special attention in subsequent calculations.

[0202] In this embodiment, the Booleanization process generates a Boolean value based on the comparison result. For example, if the average feature correlation is less than or equal to a preset recognition threshold, the Boolean value is 1; if the average feature correlation is greater than the preset recognition threshold, the Boolean value is 0.

[0203] In one specific implementation, the preset recognition threshold may be:

[0204] ;

[0205] In the formula, Indicates the recognition threshold. This indicates the setting of coefficients. represents the determination coefficient of the i-th feeder, and max represents taking the maximum value.

[0206] The similarity threshold is a standard value used to determine whether two feeders are significantly similar. If the similarity between two feeders is greater than this threshold, it indicates that their current characteristics are similar and they may have the same fault phenomenon. The identification threshold is a standard value used to determine whether a ground fault exists. It sets an appropriate threshold to determine whether the fault identification quantity is high enough, thereby identifying the line where the fault occurred.

[0207] Step S1846: Based on the Boolean quantity, calculate the cumulative difference value of each monitoring point as the fault indication quantity; wherein, the cumulative difference value is obtained by accumulating the absolute value of the difference between the average feature correlation degree of the monitoring point and the average feature correlation degree of each other monitoring point.

[0208] Understandably, faulty lines not only exhibit abnormal behavior (low average correlation), but their behavior also differs significantly from that of most normal lines. This step amplifies this inconsistency by accumulating the differences.

[0209] Therefore, in this embodiment, for each monitoring point, the cumulative difference value can be defined as: the absolute value of the difference between the average feature correlation degree of the monitoring point and the average feature correlation degree of each other monitoring point in the system, and then the sum obtained by weighted accumulation.

[0210] In a specific implementation plan, the calculation can be performed using the following formula:

[0211] ;

[0212] In the formula, βi represents the fault identification quantity of the i-th feeder. This represents the average similarity between the i-th feeder and all other feeders. The average similarity between the j-th feeder and all other feeders.

[0213] It's worth noting that by setting similarity and identification thresholds, combined with Boolean transformation, the complexity of fault location is simplified. First, by calculating the similarity between feeders and comparing it with the set threshold, feeders with similar current characteristics can be efficiently filtered out, helping to narrow down the range of possible fault locations. Boolean transformation makes fault judgment more intuitive and faster, converting continuous similarity values ​​into clear "fault" or "non-fault" states, improving the system's response speed. By setting an appropriate identification threshold, it can be ensured that only feeders that have truly experienced faults are identified, reducing the false alarm rate. Finally, the fault identifier generated based on Boolean values ​​can accurately determine the location of the fault, making fault location more efficient and precise.

[0214] In this embodiment, the faulty line can be determined based on the number of fault indicators and the different grounding methods of the distribution network.

[0215] Grounding method refers to the connection method between electrical equipment and the ground in a power distribution network. The main types include ungrounded systems, resonant grounding systems, and low-resistance grounding systems. Different grounding methods affect the fault characteristics of the power grid; therefore, the differences in grounding methods need to be considered when locating faults. The faulty line refers to the specific power line where a grounding fault occurs. In a power distribution network, locating the faulty line is crucial for rapid repair and restoration of power supply.

[0216] It should be noted that by combining fault identification quantities and recognition thresholds for different grounding methods, faulty lines can be accurately identified. Based on the electrical characteristics of different grounding methods (such as ungrounded, resonant grounding, or low-resistance grounding), the system sets different fault identification quantity judgment criteria to ensure it can handle various power grid grounding fault situations. By comparing the fault identification quantity with the recognition threshold, the specific location of the fault can be accurately determined, improving the accuracy of fault location. Especially under complex grounding methods, setting appropriate recognition thresholds and preset fault identification quantities allows the system to adapt to different types of grounding faults, reducing the possibility of misjudgment. This method is highly efficient in power grid fault location, not only accelerating the fault location process but also improving the reliability and stability of power grid operation.

[0217] In one possible implementation, the different grounding methods of the distribution network specifically include: ungrounded system, resonant grounding system, and low-resistance grounding system.

[0218] In one possible implementation, for an ungrounded system, a faulty line is identified when the number of fault indicators on the feeder exceeds an identification threshold. Here, an ungrounded system refers to a power system where the neutral point is not grounded or is insulated, and there is no direct grounding path during a fault. Due to this grounding method, the system will exhibit specific current change characteristics during a fault.

[0219] For resonant grounding systems and low-resistance grounding systems, the faulty line is identified when the number of fault indicators on the feeder is greater than the preset number of fault indicators.

[0220] Resonant grounding systems achieve grounding by introducing a small amount of capacitance and inductance into the system to adjust its resonant state. This system is typically used to reduce the impact of grounding faults, but it generates specific current characteristics when a fault occurs. Low-resistance grounding systems reduce the impact of fault currents by grounding the neutral point and using low-impedance grounding devices. This grounding method is commonly used in power systems to improve system safety and reduce fault currents. Those skilled in the art can set the preset fault indicator value according to actual conditions; this invention does not limit this.

[0221] It should be noted that precise fault location is achieved by setting fault identification quantities and corresponding thresholds for different grounding methods. Different grounding methods have different current characteristics and fault manifestations. For example, in an ungrounded system, a grounding fault does not form a direct current loop, and the change pattern of its fault identification quantity differs from other grounding methods. Furthermore, in resonant grounding and low-resistance grounding systems, the current characteristics and fault manifestations also differ. By setting appropriate fault identification quantities and identification thresholds for each grounding method, the system can accurately identify the faulty line based on the specific grounding type. In this way, the system can efficiently distinguish fault conditions under different grounding methods, avoid misjudgments, and improve the accuracy and reliability of fault location. This method not only speeds up fault location but also enhances the adaptability of the distribution network under various grounding methods, ensuring the stable operation of the power grid.

[0222] In this embodiment of the invention, three-phase currents from multiple monitoring points in the distribution network are collected, and zero-sequence current is identified as the primary characteristic of grounding faults. Then, an improved variable-step-size LMS adaptive filter is used to denoise the zero-sequence current, and combined with an elliptic high-pass filter, multiple high-frequency transient current features that trigger grounding faults are extracted. The similarity between these high-frequency transient current features is calculated using the Deng's grey relational algorithm, and a fault identifier is generated through Booleanization, simplifying the fault identification process. Finally, based on the fault identifier and the different grounding methods of the distribution network, the faulty line is identified, significantly improving the detection speed and accuracy of grounding faults, reducing fault response time, ensuring the stable operation of the distribution network, further reducing labor costs and time, and thus improving the reliability and stability of the power grid.

[0223] In one specific implementation plan, a method for locating ground faults in a distribution network based on zero-sequence current characteristics is provided.

[0224] Specifically, in a three-phase power system, zero-sequence current refers to the current caused by the unbalanced (or asymmetrical) portion of the three-phase current. Zero-sequence current typically occurs during ground faults, especially when the neutral point is ungrounded or the grounding resistance is low. The distribution network is a crucial part of the power system, responsible for transmitting electrical energy from substations to end users (such as homes and businesses). A ground fault is a fault in which a conductor (such as the neutral wire of a line) in the power system short-circuits to the ground. The distribution network ground fault location method based on zero-sequence current characteristics involves analyzing the zero-sequence current signal in the distribution network to detect specific current characteristics caused by the ground fault and using these characteristics to determine the location of the fault.

[0225] Zero-sequence current can effectively reflect the occurrence of grounding faults, especially in distribution networks with complex grounding methods. Zero-sequence current can provide more accurate fault information than traditional methods, which helps to be flexibly applied under various grounding systems, improve the stability and security of the power grid, and promote the development of distribution network automation and intelligent technology.

[0226] However, existing methods for locating ground faults in distribution networks largely rely on traditional current and voltage measurements and manual inspections, which often suffer from insufficient accuracy and response speed. In cases of complex grounding configurations or significant load fluctuations, these methods may lead to misjudgments or delays, making them ill-suited to the rapid changes and variable load conditions of power systems. Furthermore, traditional methods may struggle to handle multi-point ground faults or asymmetrical faults, failing to accurately distinguish between different fault types. To address the technical problems of existing methods relying heavily on traditional current and voltage measurements and manual inspections, which often result in insufficient accuracy and response speed, leading to misjudgments or delays and making them ill-suited to the rapid changes and variable load conditions of power systems, this invention provides a method for locating ground faults in distribution networks based on zero-sequence current characteristics, comprising:

[0227] S1: Collect the three-phase current at multiple monitoring points in the distribution network through current transformers;

[0228] S2: Determine the zero-sequence current at each monitoring point based on the three-phase current;

[0229] S3: The zero-sequence current is denoised using a variable step-size LMS adaptive filter;

[0230] S4: Based on the denoised zero-sequence current, multiple high-frequency transient current characteristics that trigger ground faults are extracted through an elliptic high-pass filter.

[0231] S5: Calculate the similarity between various high-frequency transient current features using Deng's grey relational algorithm;

[0232] S6: Based on similarity, generate a number of fault identifiers through Booleanization.

[0233] S7: Determine the faulty line based on the fault identification quantity and the different grounding methods of the distribution network.

[0234] According to an embodiment of the present invention, an electronic device is provided; please refer to... Figure 4 The electronic device in this embodiment may include one or more of the following components: a processor, a network interface, memory, non-volatile memory, and one or more application programs, wherein the one or more application programs may be stored in non-volatile memory and configured to be executed by one or more processors, and the one or more programs are configured to perform the methods as described in the foregoing method embodiments.

[0235] According to embodiments of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a computer, causes the computer to perform the method described in any of the above embodiments.

[0236] According to embodiments of the present invention, a computer program product comprising instructions is also provided, which, when executed by a computer, cause the computer to perform a method in any of the above embodiments.

[0237] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for locating grounding faults in a power distribution network, characterized in that, include: Acquire zero-sequence current signals from multiple monitoring points in the power distribution network; The zero-sequence current signal is denoised based on a preset adaptive denoising algorithm to obtain a denoised zero-sequence current signal; wherein, the adaptive denoising algorithm is configured to dynamically adjust the filtering parameters based on the error signal to suppress noise; The denoised zero-sequence current signal is processed using a preset feature extraction filter to extract high-frequency transient features; wherein, the feature extraction filter is configured to attenuate low-frequency components and retain high-frequency signals above a preset cutoff frequency; wherein, the high-frequency transient features are the features characterizing the instantaneous fluctuation characteristics of ground faults in the high-frequency signals retained by the feature extraction filter. The faulty line is determined based on the aforementioned high-frequency transient characteristics.

2. The method according to claim 1, characterized in that, The step of acquiring zero-sequence current signals from multiple monitoring points in the distribution network includes: The three-phase current at each monitoring point is collected using a pre-set current transformer; The corresponding zero-sequence current signal is calculated based on the three-phase current.

3. The method according to claim 2, characterized in that, The step of denoising the zero-sequence current signal based on a preset adaptive denoising algorithm to obtain a denoised zero-sequence current signal includes: Use the zero-sequence current signal as the input signal; Extract the signal sequence within the receptive domain from the input signal; The signal sequence within the receptive field is weighted and calculated to obtain the estimated output signal; Based on the estimated output signal and the preset expected signal, the error signal is determined; Based on the error signal, and combined with the sliding window mechanism, the weights corresponding to the signal sequence within the receptive field are updated. The above steps are executed iteratively until the error signal meets the convergence condition, thereby completing the denoising process of the zero-sequence current signal.

4. The method according to claim 3, characterized in that, The step of extracting the signal sequence within the receptive domain from the input signal includes: A continuous signal sequence is extracted from the input signal through a preset sliding window to form the receptive field region; The step of weighting the signal sequence within the receptive domain to obtain the estimated output signal includes: Based on a preset least mean square adaptive filtering algorithm, the signal sequence within the receptive field is weighted and calculated to obtain the estimated output signal; The step of updating the weights corresponding to the signal sequence within the receptive field region based on the error signal and using a sliding window mechanism includes: The updated weight vector is obtained by adding the weight adjustment amount to the original weight vector; wherein the weight adjustment amount is the product of a preset step size factor, the error signal, and the signal sequence within the receptive field.

5. The method according to claim 1, characterized in that, The step of processing the denoised zero-sequence current signal using a preset feature extraction filter to extract high-frequency transient features includes: The denoised zero-sequence current signal is filtered using a preset elliptical high-pass filter to attenuate the current components at or below the fundamental power frequency and output a high-frequency signal higher than the preset cutoff frequency. The transient fluctuation characteristics characterizing the ground fault are extracted from the high-frequency signal as the high-frequency transient features; wherein, the high-frequency transient features include at least one of the following: peak value, rising edge, pulse width, abrupt change point, frequency component, oscillation, duration, and amplitude of the zero-sequence current.

6. The method according to claim 1, characterized in that, The step of determining the faulty line based on the high-frequency transient characteristics includes: Based on the high-frequency transient characteristics of multiple monitoring points, the feature correlation degree between different monitoring points is calculated; Based on the aforementioned feature correlation, a fault indication quantity is generated to indicate the fault probability of each monitoring point; Based on the fault indication, the faulty line is determined from the plurality of monitoring points.

7. The method according to claim 6, characterized in that, The step of calculating the feature correlation degree between different monitoring points based on the high-frequency transient features of multiple monitoring points includes: The high-frequency transient characteristics of each monitoring point at each time point are normalized. For each pair of monitoring points, the grey relational coefficient of the pair of monitoring points at multiple time points is calculated based on the normalized high-frequency transient feature sequence. Based on the calculated grey relational coefficients at each time point, the average grey relational degree between the monitoring points is determined as the feature relational degree.

8. The method according to claim 6, characterized in that, The step of generating a fault indication quantity to indicate the fault probability of each monitoring point based on the feature correlation degree includes: For each monitoring point, calculate the feature correlation degree between that monitoring point and all other monitoring points, and determine the average feature correlation degree of that monitoring point; The average feature correlation degree of each monitoring point is compared with the preset identification threshold, and the comparison result is Booleanized to obtain a Boolean value; Based on the Boolean quantity, the cumulative difference value of each monitoring point is calculated as the fault indication quantity; wherein, the cumulative difference value is obtained by summing the absolute values ​​of the differences between the average feature correlation degree of the monitoring point and the average feature correlation degree of each other monitoring point.

9. An electronic device, characterized in that, include: A memory, and one or more processors communicatively connected to the memory; The memory stores instructions that can be executed by the one or more processors to cause the one or more processors to implement the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.

Images