A power distribution network fault locating method and terminal combining multi-terminal traveling wave sampling

Abstract

The application relates to a power distribution network fault locating method and a terminal combining multi-terminal traveling wave sampling. The method comprises the following steps: synchronously collecting a transient interference spectrum excited by a network disturbance singular point; performing normalization processing on the transient interference spectrum, and extracting a high-dimensional spectral feature vector therefrom; for each candidate singular point, predicting a theoretical spectral feature vector thereof through a forward propagation operator, and calculating a holographic coherence index between the theoretical spectral feature vector and an actually observed spectral feature vector, so as to iteratively reconstruct a singular point holographic data body representing a global fault probability distribution; performing image decoding analysis on the singular point holographic data body, and determining the accurate position of the network disturbance singular point by searching for a global maximum value point. The application converts the fault locating problem into an image reconstruction and understanding problem, utilizes complete information of a traveling wave signal instead of single time information, and significantly improves the positioning accuracy and robustness.

Description

Technical Field

[0001] This application relates to the field of power system technology, and in particular to a method and terminal for fault location in distribution networks that combines multi-terminal traveling wave sampling. Background Technology

[0002] As a bridge connecting high-voltage transmission networks and end users, the safe and stable operation of distribution networks is crucial for ensuring social production and residents' lives. However, due to the complex structure, numerous branches, and wide distribution of distribution network lines, and their susceptibility to external environmental factors (such as severe weather and external damage), the failure rate is relatively high. When a failure occurs, quickly and accurately locating the fault point is a key technical challenge for shortening power outage time, improving power supply reliability, and reducing operation and maintenance costs.

[0003] Traditional methods for fault location in distribution networks mainly include the impedance method based on power frequency information and the traveling wave method based on transient signals. The impedance method estimates the fault distance by measuring the voltage and current after the fault and calculating the line impedance. This method is effective in radial networks with simple topologies, but in complex distribution networks with multiple branches and ring network structures, the location accuracy is difficult to guarantee due to the influence of multiple power sources, load fluctuations, and transition resistance.

[0004] Traveling wave (TW) method uses transient traveling wave signals generated at the instant of a fault for location. Its basic principle is that the traveling wave generated at the fault point propagates along the line to both sides at near the speed of light. By installing monitoring devices at different locations on the line to capture the arrival time of the traveling wave, the location of the fault point can be calculated using the time difference of arrival and the known wave speed. The TW method based on two-end measurement is a typical example, but it requires the fault point to be located between two monitoring devices, making it ineffective for faults on branch lines. To cover more complex networks, multi-end TW location technology has been developed. However, most existing multi-end TW location technologies still rely on the precise determination of the arrival time of the "wavefront." In actual distribution networks, due to background noise, signal attenuation, dispersion effects, and interference from multipath reflection and refraction, the wavefront of the traveling wave is often distorted, making the identification of the wavefront arrival time very difficult. This is the main bottleneck limiting its location accuracy. Furthermore, these methods only utilize the single timestamp information in the traveling wave signal, ignoring the information carried by the traveling wave waveform itself that reflects the nature of the fault and the state of the propagation path, resulting in a serious lack of information utilization.

[0005] With the development of signal processing and artificial intelligence technologies, some studies have attempted to analyze traveling wave signals more deeply. However, most of these studies remain focused on fault type identification at a single measurement point, failing to effectively combine information from multiple measurement points to solve spatial location problems from a global and systematic perspective. In particular, how to fuse one-dimensional time-series signals collected from multiple spatially distributed sensors into a high-confidence, globally optimal solution regarding the spatial location of the fault source remains an unresolved technical challenge. Summary of the Invention

[0006] The purpose of this application is to provide a distribution network fault location method, terminal and computer-readable storage medium that combines multi-terminal traveling wave sampling, aiming to solve the technical problems mentioned in the background art, such as excessive reliance on wavehead identification, low information utilization, and insufficient location accuracy and robustness in complex networks in existing traveling wave location methods.

[0007] Firstly, this application provides a method for fault location in distribution networks that combines multi-terminal traveling wave sampling. This method uses multiple synchronous sampling array nodes deployed in the distribution network to synchronously acquire transient interferometric spectra excited by singularities in network disturbances. This step utilizes distributed high-speed acquisition terminals to completely capture the information-rich traveling wave signal generated instantaneously during the fault. Subsequently, the method normalizes the acquired transient interferometric spectra and extracts a spectral feature vector from each transient interferometric spectrum.

[0008] This application utilizes deep feature engineering to map the original one-dimensional time-domain waveform to a high-dimensional feature space, thereby enabling a more comprehensive and robust characterization of the signal's intrinsic properties and eliminating dependence on a single wavefront time point. Next, within a pre-defined waveguide network medium model, for each candidate singularity in the model, the method obtains its corresponding theoretical spectral feature vector. Based on this theoretical spectral feature vector and the actually observed spectral feature vector, it obtains the holographic coherence index of the candidate singularity to generate a singularity holographic data volume. This application creatively transforms the fault location problem into an image reconstruction problem. By traversing all possible fault locations and calculating their matching degree with global observation data, a global fault probability distribution map is generated.

[0009] Finally, the method decodes the singularity holographic data volume to determine the location of the network disturbance singularity in the distribution network. This step, by applying image analysis technology, accurately resolves the fault point from the probability distribution map, ensuring the global optimality of the location result. The method proposed in this application significantly improves the accuracy, robustness, and adaptability to complex networks of fault location by constructing a complete technical closed loop from multi-dimensional signal acquisition, deep feature extraction, global model matching to image decoding.

[0010] In one possible implementation of the first aspect, the process of synchronously acquiring transient interferometric spectra includes: multiple synchronous sampling array nodes achieving nanosecond-level time synchronization based on a unified high-precision clock source (such as a GPS / BeiDou timing module), ensuring the consistency of the time reference for subsequent data processing. Furthermore, after a network disturbance singularity occurs, these nodes are efficiently activated by preset triggering logic (such as a sudden change in the amplitude or steepness of the traveling wave signal), and each node captures and stores a data frame of a preset length as a transient interferometric spectrum, thereby completely recording the key information of the traveling wave.

[0011] In one possible implementation of the first aspect, the process of extracting the spectral feature vector includes: performing a time-frequency joint transform (such as wavelet transform) on the transient interferometric spectrum to generate a two-dimensional time-spectrum. This transform reveals the dynamic characteristics of the signal frequency changing over time. Then, at least one time-frequency joint feature (such as texture features or energy concentration) is extracted from this two-dimensional time-spectrum and these features are incorporated as components of the spectral feature vector. This elevates one-dimensional signal analysis to two-dimensional image analysis, greatly enriching the dimensionality and representational capabilities of the features.

[0012] In one possible implementation of the first aspect, the process of obtaining the holographic coherence index includes: using a forward propagation operator that, based on the position of the candidate singularity in the waveguide network medium model, can physically or empirically predict the theoretical spectral feature vectors that should be observed at each synchronous sampling array node when a fault occurs at that location. Then, the candidate holographic coherence index is obtained by combining (e.g., calculating the reciprocal of the weighted distance) the vector distances or similarities between these theoretical spectral feature vectors and the actually observed spectral feature vectors.

[0013] In one possible implementation of the first aspect, the process of decoding the singularity holographic data volume includes: searching for the candidate singularity with the globally maximum holographic coherence index in the scalar field represented by the singularity holographic data volume and defined on the distribution network topology. Then, the position of the candidate singularity is directly determined as the final position of the network disturbance singularity.

[0014] Secondly, this application provides a method for fault location in distribution networks, which aims to address the problem of decreased location accuracy caused by the mismatch between the guided wave network dielectric model and the actual physical network state. This method periodically injects a calibration pulse signal with known characteristics into the distribution network, and multiple synchronous sampling array nodes synchronously record its response, i.e., the calibration transient interferogram. Then, the calibration spectrum feature vector is extracted and compared with the theoretical calibration spectrum feature vector predicted by the current guided wave network dielectric model to obtain the model error. Finally, based on this model error, one or more key parameters in the guided wave network dielectric model (such as the line traveling wave velocity) are adaptively updated. This active detection and feedback calibration mechanism enables the location system to track the gradual changes in line parameters in real time, ensuring continuous high-precision location capability, and has significant engineering application value.

[0015] In one possible implementation of the second aspect, the updated model parameters are specifically the traveling wave velocity of the distribution network lines.

[0016] In one possible implementation of the second aspect, the parameter update process is performed by an optimization algorithm (such as gradient descent), which iteratively adjusts the model parameters to systematically minimize the difference between theoretical predictions and actual measurements, thereby achieving automated and intelligent model calibration.

[0017] Thirdly, this application provides a power distribution network fault location terminal, which integrates functional modules for implementing the method described in the first aspect, including: an acquisition module responsible for synchronously acquiring transient interference spectra; a feature extraction module responsible for converting the original waveform into a high-dimensional feature vector; a reconstruction module responsible for generating the core singularity holographic data volume; and a decoding module responsible for parsing the final fault location from the data volume. This terminal provides the hardware foundation for the engineering implementation of the method.

[0018] Fourthly, this application provides a computer-readable storage medium on which a computer program stored, when executed by a processor, can implement any of the aforementioned methods. This makes it possible to upgrade existing hardware devices through software or deploy the technical solution of this application on a general-purpose computing platform. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a core fault location method according to an embodiment of this application.

[0020] Figure 2 This is a schematic flowchart of a model adaptive calibration mechanism according to another embodiment of this application.

[0021] Figure 3This is a functional block diagram of a power distribution network fault location terminal according to an embodiment of this application. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0024] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0025] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0026] Example 1

[0027] This embodiment provides a distribution network fault location method combining multi-terminal traveling wave sampling. This method reconstructs the distribution network fault location problem from traditional one-dimensional time signal measurement into an image reconstruction and understanding problem based on multi-dimensional feature fusion and global field theory analysis. The method abstracts the distribution network as a waveguide network medium and considers fault events as network disturbance singularities (hereinafter referred to as singularities or fault points) generated in this medium. A synchronous sampling array is used to capture traveling wave signals excited by the singularities; these signals are defined as transient interferometric spectra. The core of the method lies in generating a singularity holographic data volume through a traveling wave field-of-view reconstruction algorithm, and finally accurately resolving the location of the disturbance singularity from it using image decoding technology.

[0028] like Figure 1 As shown, the method mainly includes the following steps:

[0029] S100: Through multiple synchronous sampling array nodes deployed in the distribution network, the transient interference spectrum excited by the network disturbance singularity is collected synchronously.

[0030] To achieve distortion-free capture of high-frequency transient traveling wave signals generated by network disturbance singularities, multiple synchronous sampling array nodes need to be deployed at key locations in the distribution network, such as substation busbars, feeder switches, and major branch nodes. These nodes together constitute a distributed, large-aperture observation array. Each node is equipped with a high-speed data acquisition unit and a high-precision clock synchronization unit.

[0031] The core of the data acquisition unit is an analog-to-digital converter (ADC), whose sampling rate is set to a sufficiently high value to satisfy the Nyquist theorem requirements for wideband traveling wave signals. For example, the sampling rate... It can be set to 10MHz, which means every A sample of the instantaneous voltage or current value on the line is taken every microsecond (μs). This high sampling rate ensures that the waveform details of the traveling wave signal, especially the wavefront portion that determines its high-frequency characteristics, can be accurately digitized and recorded.

[0032] To ensure that all nodes in the array have a unified time reference, high-precision clock synchronization units need to be installed at each node. In a preferred embodiment, each node integrates a Global Navigation Satellite System (GNSS)-based timing module, such as a GPS or BeiDou module. These modules can receive satellite signals and extract time information accurate to the nanosecond (ns) level, typically achieving a time synchronization accuracy better than 100ns. Alternatively, in a network with suitable communication capabilities, the same high-precision time synchronization can be achieved through network messages using the IEEE 1588 Precision Time Protocol (PTP).

[0033] When the distribution network system is operating normally, the transient recording units of each sampling node are in standby monitoring mode. To enable instantaneous response when disturbance singularities occur and to avoid recording massive amounts of invalid data, the system employs a triggering mechanism. This triggering logic can be based on a combination of one or more criteria. A common criterion is a sudden change in the amplitude of a traveling wave signal. For example, a voltage amplitude trigger threshold can be set. When the absolute change of the instantaneous value of the monitored line voltage Exceed At this time, the triggering logic is activated. Another, more robust criterion is based on the steepness of the waveform, i.e., the rate of change of voltage or current over time. or By calculating the difference between consecutive sampling points and applying a steepness threshold... By comparison, the arrival of the traveling wave front can be captured more reliably because it usually has a higher rate of change than the power frequency background noise.

[0034] Once the triggering logic of any node is activated, that node immediately begins storing the high-speed sampled data stream into its internal circular buffer. Understandably, in order to fully record the entire transient process, data recording does not begin at the moment of triggering, but includes a period of time before the trigger, which is called pre-trigger recording.

[0035] For example, the total length of a data frame It can be set to 2 milliseconds (ms). The data recording window can be set to start from the trigger time. Starting 0.5ms before the trigger and continuing until 1.5ms after, a captured transient interferometric spectrum thus encompasses the complete process from the system's normal state to the arrival of the traveling wave, and then to subsequent oscillations and decay. This data, with a precise GNSS timestamp, is from... The digital sequence composed of sampling points is a raw transient interference spectrum.

[0036] Finally, all triggered nodes package their recorded transient interferometric spectrum data, along with metadata such as the precise timestamp of the trigger moment, node ID, and line ID, and upload it to a central processing unit via a communication network (such as 5G, industrial Ethernet, or fiber optic network). The central processing unit can be a server deployed at the main station or a cloud computing platform. This completes the acquisition of synchronous array observation data containing global spatial information, triggered by a network disturbance singularity.

[0037] For example, suppose that in a fault, all three nodes (Node A, Node B, and Node C) in the array are successfully triggered. The central processing unit receives three sets of data. Taking the data from Node A as an example, it can be represented as a data structure containing: Node_ID:"A", Timestamp_UTC:"2025-08-2813:22:05.123456789", and a one-dimensional floating-point array WaveformData of length 20000:[0.01,0.02,...,15.78,14.23,...,-0.05]. Each element of the WaveformData array represents the voltage value (in kV) of a sampling point. This array is the transient interferometric spectrum captured by Node A. The data structures of the other nodes are similar, but their Timestamp_UTC and WaveformData will differ due to their different positions in the network. This set of multiple such data structures is the final output of step S100. Each sampling point, like an interference fringe in a hologram, has limited meaning on its own, but together they contain all the information needed to reconstruct the object. The purpose of this step is to acquire these electromagnetic interference fringes generated by the fault.

[0038] S200. Normalize the multiple transient interference spectra and extract a spectral feature vector from each of the transient interference spectra.

[0039] After S100 collects the raw transient interferometric spectrum data, S200 transforms these raw waveform data, which are noisy and have individual differences, into a standardized, information-density mathematical representation that is friendly to subsequent model matching algorithms, namely, the spectral feature vector.

[0040] First, normalization preprocessing of the transient interferometric spectrum is performed. Different sampling nodes may use different types of sensors (such as capacitive voltage sensors, Rogowski coils, etc.), and their installation locations may have differences in line impedance and grounding conditions, leading to inconsistencies in amplitude and baseline of the acquired signals. To eliminate these systematic deviations unrelated to the fault itself, normalization is necessary. This process mainly includes two sub-steps: baseline correction and amplitude normalization. In the captured data frames, the initial portion (pre-trigger recording portion) typically corresponds to the system's normal operating state; theoretically, its signal should fluctuate around zero. By calculating the average value of this pre-trigger data, the DC bias or baseline drift of that channel can be obtained. Subtracting this average value from the entire data frame completes the baseline correction. To eliminate the influence of different sensor gains, the amplitude of each data frame can be normalized. For example, maximum value normalization can be used, which involves dividing the value of each sampling point in the data frame by the maximum absolute value of all sampling points in that data frame. After this processing, the amplitudes of all transient interference spectra are scaled to the range of [-1, 1], making them numerically comparable.

[0041] After normalization, a high-dimensional spectral feature vector is extracted from each normalized transient interferometric spectrum. This application does not rely on any single, easily disturbed feature (such as wavefront arrival time), but rather extracts a series of features that can describe waveform characteristics from different perspectives and dimensions, and combines them into a vector to form a robust and comprehensive identifier for the transient interferometric spectrum. The spectral feature vector mainly comprises time-domain morphological features, frequency-domain structural features, and joint time-frequency features.

[0042] Temporal morphological characteristics are direct descriptions of the geometric shape and statistical properties of a waveform along the time axis. The temporal morphological characteristic set defined in this application may include peak factor, pulse width, wavefront rise time, and zero-crossing rate. The peak factor is defined as the ratio of the waveform's peak value to its effective mean square value (RMS), reflecting the waveform's impulse characteristics. The pulse width can be defined as the time interval between the waveform amplitude first exceeding 50% of its peak value and the last falling below 50% of its peak value, reflecting the duration of the transient process. The wavefront rise time is defined as the time required for the waveform to rise from 10% to 90% of its peak value; it is a key indicator reflecting the wavefront steepness. The zero-crossing rate is the number of times the waveform signal crosses zero within a time window, indirectly reflecting the signal's main frequency components.

[0043] For example, the Crest Factor (CF) can be precisely defined as the ratio of the maximum absolute value of the signal to the effective value (RMS) of the signal:

[0044]

[0045] The pulse width can be defined as the time difference between the moment when the signal amplitude first exceeds 50% of the peak value and the moment when it last falls below 50% of the peak value.

[0046] Frequency domain structural features are obtained by transforming a signal from the time domain to the frequency domain, revealing the signal's frequency composition and energy distribution. This application obtains the spectrum of a transient interferometric spectrum by applying a Fast Fourier Transform (FFT). From this spectrum, a series of frequency domain structural features can be extracted. These features may include the dominant band energy proportion, spectral entropy, and spectral centroid. The dominant band energy proportion is defined as the ratio of the total energy within a critical frequency band (e.g., 500kHz-2MHz) to the total signal energy, reflecting specific frequency components related to the fault type. Spectral entropy is an indicator used to measure spectral uncertainty or complexity; the flatter the spectrum, the higher its entropy value. The spectral centroid, or the center frequency of the spectrum, is the center of mass of the spectrum; the richer the high-frequency components, the higher the value of the spectral centroid.

[0047] Joint time-frequency features are used to capture the non-stationary characteristics of the dynamic evolution of signal frequency components over time. To extract these features, this application employs continuous wavelet transform (CWT) to analyze transient interferometric spectra. CWT uses a mother wavelet function (e.g., the Morlet wavelet) and convolves it with the original signal by scaling and translating it. The result is a two-dimensional coefficient matrix, with one dimension representing time and the other representing scale (inversely proportional to frequency). The values ​​of the matrix elements represent the energy intensity of the signal at that time and scale. This two-dimensional matrix can be visualized as a scalogram, where the horizontal axis represents time, the vertical axis represents frequency, and the color intensity represents energy level. In this application, the scalogram is directly considered as an image that needs to be understood and analyzed. Various joint time-frequency features can be extracted from this image. These features may include image texture features, such as obtaining texture descriptors like contrast, correlation, energy, and homogeneity by calculating the gray-level co-occurrence matrix (GLCM) of the scalogram. These texture descriptors quantify the distribution pattern of traveling wave energy in the time-frequency plane. It can also include energy concentration, such as calculating the proportion of the area of ​​the region with the most concentrated energy to the total area, which can reflect the transient characteristics of the signal.

[0048] For example, the image texture features can be obtained by calculating the gray-level co-occurrence matrix (GLCM) of the time-space spectrogram. Specifically, the GLCM with a distance of 1 and orientations of 0° and 90° can be calculated, and its contrast index can be extracted as a feature. The calculation formula is as follows:

[0049]

[0050] Where P(i,j) is the probability value at the corresponding position in GLCM.

[0051] Finally, all the extracted and quantized time-domain features, frequency-domain features, and time-frequency joint features are concatenated into a single, fixed-dimensional column vector according to a predefined order. This vector is the final mathematical expression of the original transient interferometric spectrum—the spectral feature vector. For example, a spectral eigenvector can be constructed into a 10-dimensional vector. The first four components are time-domain features, the next three are frequency-domain features, and the last three are joint time-frequency features. The output of step S200 is to generate such a spectral feature vector for each transient interferometric spectrum.

[0052] For example, continuing the example in S100, feature extraction is performed on the normalized WaveformData (an array of length 20000) captured by Node A.

[0053] Time-domain feature calculation:

[0054] The calculated peak factor is 3.5.

[0055] The calculated pulse width is 15.2 μs.

[0056] The calculated wavefront rise time is 1.8 μs.

[0057] The calculated zero-crossing rate is 1.2 MHz.

[0058] Frequency domain feature calculation:

[0059] Perform an FFT on the WaveformData to obtain the spectrum.

[0060] The calculated energy percentage within the [500kHz, 2MHz] frequency band is 0.68.

[0061] The calculated spectral entropy is 4.2.

[0062] The calculated spectral centroid is 1.1 MHz.

[0063] Time-frequency joint feature calculation:

[0064] Perform CWT on WaveformData, using Morlet mother wavelet, to generate a time-spectrum matrix of, for example, 256x1024.

[0065] Treat the matrix as a grayscale image, calculate its GLCM, and extract the texture contrast from it, which is 0.85.

[0066] The calculated energy concentration degree is 0.75.

[0067] Another texture feature is calculated, for example, homogeneity is 0.92.

[0068] Vector construction:

[0069] The 10 calculated values ​​are combined into a spectral feature vector in a predetermined order.

[0070] Finally, the transient interference spectrum of Node A is expressed as .

[0071] Similarly, the same processing procedure is performed on the transient interferometric spectra of Node B and Node C to obtain their respective spectral eigenvectors. and This is a set consisting of multiple observed spectral eigenvectors. This is the final output of step S200, serving as the core input for the subsequent traveling wave field-of-view reconstruction algorithm. Thus, through the above transformation, the original waveform data, which is difficult to compare directly, becomes a mathematical object that can be used to measure distance and similarity in multi-dimensional space.

[0072] It should be noted that the specific calculation methods for the various time-domain, frequency-domain, and time-frequency joint features mentioned in this step are all well-known techniques in the fields of digital signal processing and image processing. For example, statistical quantities such as peak factor and zero-crossing rate have standard mathematical definitions; the calculation of Fast Fourier Transform (FFT), Continuous Wavelet Transform (CWT), and Gray-Level Co-occurrence Matrix (GLCM) can all be implemented by calling standard functions in existing, mature scientific computing libraries (such as MATLAB®'s Signal Processing Toolbox or Python™'s SciPy, PyWavelets, Scikit-image, etc.). Based on the explicit feature definitions given in this application (such as the specific aperture definition of pulse width), those skilled in the art can easily reproduce the entire feature extraction process described in this step without any creative effort using these common tools. Therefore, the internal implementation details of these standard algorithms will not be elaborated here.

[0073] S300. In a preset waveguide network medium model, for each candidate singularity in the model, obtain its corresponding theoretical spectral feature vector, and based on the theoretical spectral feature vector and the spectral feature vector, obtain the holographic coherence index of the candidate singularity to generate a singularity holographic data volume.

[0074] After S200 converts the transient interferometric spectra of all observation stations into spectral eigenvectors, S300 performs wavefield reconstruction. The goal of this step is to inversely deduce the most probable location of the network perturbation singularity in space based on the actually observed eigenvectors. This is achieved by constructing a global holographic coherence scalar field to replace the traditional triangulation method based on local time information. This scalar field, once completed, constitutes the singularity holographic data volume.

[0075] This step begins by constructing a digital model of the waveguide network medium. This model forms the physical basis for all subsequent calculations. In one embodiment, this model is constructed as a mathematical graph structure. ,in It is a collection of nodes, representing all key physical locations such as substations, switches, and branch points in the distribution network; It is a set of edges, representing the physical line segments connecting these nodes. Each node in the graph... Each edge in the diagram is assigned its precise geospatial coordinates or mileage position within the route. They were all endowed with key physical properties, primarily including the length of the line. And the propagation speed of traveling waves on this line segment In the initial stage, wave speed A theoretical or empirical value can be set based on the line type (e.g., overhead line, cable) and its insulation medium (e.g., approximately for overhead lines). This graphical model fully describes the topological path of the traveling wave propagating in the power distribution network medium (m / s).

[0076] To perform subsequent calculations, this continuous waveguide network medium model needs to be spatially discretized. That is, all possible fault locations in the network are divided into a finite, dense candidate set. For example, each line segment in the network can be discretized according to a preset spatial resolution. The area is divided into segments (e.g., 10 meters). Each segment is defined as a candidate singularity. In this way, the entire power distribution network is represented as a discrete set consisting of tens of thousands of candidate singularities. The subsequent reconstruction algorithm essentially involves finding the most likely candidate from this vast candidate set through a global evaluation.

[0077] Next, we need to define the forward propagation operator. The operator is used to characterize: assuming the network perturbation singularity occurs at any given candidate singularity. At that point, then, in the first place What kind of signal should be observed at each node of the synchronous sampling array? More specifically, the input to this operator is the candidate singularity. The location is given by the output at each observation node. At this point, the corresponding theoretical spectral eigenvector The operator F is a function consisting of multiple independent scalar evolution functions Fi. K The vector function formed. Specifically, if the spectral eigenvector V contains n eigencomponents. Then the forward propagation operator F also consists of n corresponding evolution functions:

[0078]

[0079] Each evolution function The function is to determine the candidate singularity. Propagation distance between observation node i and the reference characteristic value of the source signal. To predict the theoretical eigenvalues ​​that should be observed at observation node i. .Right now:

[0080]

[0081] This decoupled structure allows for independent modeling of the physical properties of each feature component. The following sections will illustrate these evolution functions using several typical feature components as examples. The specific construction method.

[0082] In one embodiment, a standard reference feature vector describing typical fault source signals is first defined. This vector can be obtained through extensive electromagnetic transient simulations or analysis of historical fault data. Then, for any candidate singularity... and observation nodes First, use a graphical model Calculate the shortest propagation distance between them Subsequently, the predicted feature vectors The individual components can be calculated using the following set of exemplary evolution formulas:

[0083] Evolution of temporal morphological features (based on) (Taking wave rise time as an example): During the propagation of a traveling wave, due to dispersion, its wavefront gradually becomes flatter, meaning the rise time increases. This change can be approximated by a linear model:

[0084]

[0085] in, This is the source wavefront rise time of the reference, which can be exemplarily taken as a value of... . This is the dispersion coefficient of the line, representing the increase in rise time of a traveling wave for every kilometer it travels. This coefficient is related to the specific parameters of the line, and in a preferred embodiment, it can be taken as [value missing]. .

[0086] Frequency domain structural feature evolution (with Taking the dominant frequency band energy percentage as an example): The high-frequency components of a traveling wave attenuate faster than the low-frequency components during propagation; this is a fundamental physical law governing electromagnetic wave propagation in lossy media. Therefore, the energy percentage of the dominant frequency band (usually high frequency) decreases exponentially with distance.

[0087]

[0088] in, It is the proportion of the main frequency band energy of the source signal, for example, it can be set to . It is the overall attenuation coefficient of the signal within this frequency band, and can be exemplarily taken as a value of .

[0089] Time-frequency joint feature evolution (with (Taking texture contrast as an example): Signals typically have the most complex time-frequency structure at the source, where the texture contrast of the time-frequency spectrum is highest. As the propagation distance increases, multipath reflection and dispersion cause the time-frequency structure of the signal to become blurred and smoother, leading to a decrease in texture contrast. This change can be described by a bounded exponential decay function:

[0090]

[0091] in, It is the reference texture contrast of the source signal, for example . It is the minimum texture contrast of the signal after it has traveled indefinitely (i.e., when only background noise remains), for example . It is the smoothing coefficient of the time-frequency structure, and can be exemplarily taken as a value of .

[0092] It should be emphasized that the above evolution formulas for wavefront rise time, dominant frequency band energy proportion, and texture contrast are merely preferred embodiments. Those skilled in the art can construct the corresponding evolution functions based on the physical meaning of different feature components through one or more of the following methods. :

[0093] Analyze the intrinsic relationship between specific characteristic components and the physical processes of traveling wave propagation (such as attenuation, dispersion, and multipath reflection). For example, for the peak factor ( Its value is mainly affected by the signal amplitude attenuation, therefore an exponential attenuation model related to the attenuation coefficient can be constructed; for pulse width ( Its value is mainly affected by the dispersion effect, and a model can be constructed that increases linearly or nonlinearly with distance.

[0094] Extensive simulations and statistical analyses are performed using electromagnetic transient simulation software (such as PSCAD / EMTP) or historical fault data. By setting faults at different locations, the numerical changes of each spectral characteristic component are recorded as the traveling wave propagates to different observation points. Then, standard curve fitting techniques (such as least squares method, polynomial fitting) or machine learning regression algorithms (such as support vector regression, neural networks) are used to establish the characteristics of each component. With transmission distance The functional mapping relationship between them is used to obtain its empirical evolution function. .

[0095] Based on the above method, those skilled in the art can systematically construct a complete forward propagation operator F for all components of any given spectral feature vector V.

[0096] It should be noted that the specific evolution formulas provided in the embodiments of this application are a preferred implementation based on physical laws and empirical summaries, but the present invention is not limited thereto. Those skilled in the art can select or develop appropriate modeling methods to construct the forward propagation operator according to specific application scenarios and available data resources.

[0097] After defining the forward propagation operator Then, we can begin calculating the holographic coherence index for each candidate singularity. This index quantifies when a failure occurs. This assumption demonstrates global consistency or coherence with all actual observation data. The calculation process is as follows:

[0098] For a specific candidate singularity :

[0099] Using the forward propagation operator For each observation node (For example, nodes A, B, and C) calculate their theoretical spectral eigenvectors: , , .

[0100] For each observation node Calculate the spectral eigenvectors actually observed. Compared with the theoretically predicted spectral eigenvectors The dissimilarity between them. This dissimilarity can be expressed by a vector distance function. To measure.

[0101] For example, a weighted squared Euclidean distance can be used: .in It is The diagonal weight matrix can assign higher weights to the more reliable and important components in the eigenvector.

[0102] The dissimilarity of all observed nodes is weighted and summed to obtain a total mismatch degree. The weights here The signal-to-noise ratio (SNR) of each observation node can be set, and the higher the SNR, the greater the contribution of the signal in the final decision.

[0103] Finally, the total mismatch is converted into a holographic coherence index through an inverse function (e.g., the reciprocal). .in It is a very small positive integer used to prevent the denominator from being zero. It is defined in this way. The higher the value, the more likely the fault occurred. The more closely the "point" assumption matches the global observation data, the better.

[0104] Optionally, to optimize the performance of the distance metric, the aforementioned weight matrix... This can be systematically determined based on the statistical properties of the features, rather than being set empirically. In one embodiment, a large amount of background noise data during normal operation and typical fault simulation data can be collected first. Then, for each feature component in the spectral feature vector... Calculate the "discrimination" of each feature (from 1 to 10) in distinguishing between faults and noise. This discrimination can be quantified by calculating the ratio of inter-class divergence to intra-class divergence. Finally, feature components with higher discrimination are assigned larger weight values. For example, a matrix... diagonal elements It can be set to be the same as the first The discriminative power of each feature is directly proportional to its individual characteristics. For example, if a joint time-frequency feature is found... Its distinguishability is much higher than that of time-domain features. Then you can set ,and This makes the setting of weights based on evidence, thereby enhancing the integrity and robustness of the technical solution.

[0105] By traversing all candidate singularities in the network model (from arrive And calculate the corresponding holographic coherence index for each point. This yields a scalar field defined over the entire distribution network topology space. This is formed by a set The resulting dataset is the singularity holographic data volume defined in this invention. Conceptually, this data volume is equivalent to a three-dimensional, global fault probability density map, whose "brightness" is... The value of .

[0106] For example, continuing the previous example, the assumed distribution network model has been discretized into 10,000 candidate singularities. To calculate two of the points. and The holographic coherence index. Assuming the real fault point occurs... nearby.

[0107] calculate :

[0108] Calling the forward propagation operator The theoretical prediction vector is obtained as follows:

[0109]

[0110] (The prediction vectors for other nodes B and C are calculated similarly...)

[0111] Calculate the dissimilarity of each node (assuming...) (The identity matrix, i.e., unweighted)

[0112]

[0113] Assuming the calculation yields , .

[0114] Calculate the total mismatch (assuming the weights of each node). ):

[0115]

[0116] Calculate the holographic coherence index (assuming...) ):

[0117]

[0118] calculate (This is a location far from the actual point of failure):

[0119] Calling the forward propagation operator Because of incorrect location assumptions, the predicted vectors will deviate significantly from the actual observed values.

[0120]

[0121] Calculate dissimilarity:

[0122]

[0123] Assuming other nodes also calculate a large dissimilarity, we get , .

[0124] Calculate the total mismatch:

[0125]

[0126] Calculate the holographic coherence index:

[0127]

[0128] Through this calculation process, we can see the candidate points that correspond to the actual fault points. A very high coherence index (7.429) was obtained, while the actual fault location was far away. This yields a very low index (0.000745). After performing this calculation on all 10,000 candidate points, the output of step S300 is a singularity holographic data volume containing 10,000 coherence index values. This data volume, when visualized, will... A bright "spot" forms nearby, while other areas are relatively dim, thus providing an intuitive basis for the final location.

[0129] S400. Decode the singularity holographic data volume to determine the location of the network disturbance singularity in the power distribution network.

[0130] After the S300 reconstructs the singularity holographic data volume, the S400 performs the final decoding on this data volume containing global position information to extract the unique and precise position coordinates of the network perturbation singularity.

[0131] The singularity holographic data volume itself, as a scalar field defined on a one-dimensional topology (line) of the power distribution network, can be regarded as one or more one-dimensional "images". The decoding process is to analyze these images and find the most significant feature points.

[0132] In a preferred embodiment, the core decoding operation is searching for the global maximum value point within the singularity holographic data volume. This means that the processing unit needs to traverse all candidate singularities generated by S300. and its corresponding holographic coherence index And find that makes The candidate singularity with the largest value This process can be expressed mathematically as follows:

[0133]

[0134] in, This is the location ultimately determined to be the singularity of the network disturbance. Due to the holographic coherence index... The construction method ensures that the higher the value, the better the agreement between the hypothesis that the point is the source of the fault and all observed data. Therefore, the location corresponding to this global maximum point is the optimal estimate in the global scope based on all available information. This global search strategy fundamentally avoids the location error caused by local optima that may occur in traditional methods.

[0135] To improve decoding efficiency, especially when the number of candidate singularities is very large, optimized search algorithms can be used instead of exhaustive traversal. For example, a global search can be performed first at a coarser spatial resolution to quickly locate a "highlighted region" containing the maximum value. Then, within this smaller region, a second, finer search can be performed at a higher spatial resolution, thus achieving a balance between computational efficiency and positioning accuracy.

[0136] Besides determining the fault location, this step can also perform further analysis of the image to extract more useful information, such as evaluating the confidence level of the localization results. An important evaluation metric is peak sharpness. If the holographic coherence index reaches its maximum value... A sharp peak forming near the point and rapidly attenuating in areas far away indicates a high degree of uniqueness and certainty in the location result, signifying high confidence. Conversely, a flat peak or multiple secondary peaks with similar values ​​may indicate poor signal quality or model uncertainty, resulting in lower confidence. Peak sharpness can be quantified, for example, by calculating the second derivative (or difference) between the maximum value and its neighbors. A high-confidence location result can be automatically confirmed by the system, while a low-confidence result can trigger an alarm, prompting maintenance personnel to conduct manual analysis in conjunction with other information.

[0137] For example, the quantization calculation of peak sharpness can be performed as follows: assuming It is the point of global maximum value, and its coherence index is In its two adjacent candidate singularities along the route topology. and At this point, the coherence indices are respectively and The peak sharpness S can then be defined as the ratio of the difference between the center point and the average values ​​on both sides, approximating the second difference:

[0138]

[0139] A high S value (e.g., greater than 0.8) indicates a very sharp peak and a high confidence level in the localization result.

[0140] Furthermore, this step can also perform a preliminary inversion of the properties of the perturbation singularity. This is after the final fault location has been determined. Then, this position can be substituted back into the forward propagation operator defined in S300. Thus, the most theoretical predicted spectral feature vectors at each observation node are obtained when a fault occurs at that location. By comparing the optimal prediction vector with the actual observed vector in detail... The residuals between (i.e.) This allows us to obtain clues about the characteristics of the fault itself. For example, if the residual vector exhibits a specific pattern in the components related to high-frequency energy, it may correspond to a specific fault type (such as metallic grounding or arc grounding). By establishing a mapping library between residual patterns and fault types (which can be learned through simulation or historical data), we can provide a preliminary judgment of the fault type while locating the fault, thus providing more valuable information for subsequent fault handling and repair.

[0141] Continuing with the example of S300, the central processing unit receives 10,000... The holographic data volume of the singularity.

[0142] Global optimum search: The processing unit executes a loop, comparing all The value of . After the traversal is complete, it is found that when hour, It is the largest of all values.

[0143] Location determined: Therefore, the system will identify candidate singularities. The location was determined as the final fault location. Assume... If it is defined in the database as "Feeder No. 3, 1.25 kilometers away from the switch cabinet", then this is the final location report output to the operation and maintenance personnel.

[0144] Confidence assessment: The system then analyzes... The surrounding coherence index value. For example, it found , And points further away, such as It has dropped to 0.1. This rapid decline in the index indicates a very sharp peak. Based on the preset sharpness calculation formula, the system obtains a confidence score as high as 0.95 (out of 1).

[0145] Attribute inversion: The system calculates the residuals between the optimal predicted vector and the actual observed vector. It is found that the residuals exhibit a systematic bias at all three nodes: an excessively high "spectral entropy" component and an excessively large "wavefront rise time" component. By querying the fault knowledge base, the system matches this residual pattern to a high correlation with "high-impedance arc grounding fault".

[0146] Ultimately, the complete report output by S400 was: "A fault occurred on feeder 3, 1.25 km from the switchgear. Location confidence level: 95%. Preliminary fault type is: high-impedance arc grounding."

[0147] Example 2

[0148] In the method described in Embodiment 1, the traveling wave field-of-view reconstruction algorithm in step S300 highly depends on an accurate waveguide network dielectric model. However, in actual operating environments, the physical parameters of a distribution network are not static. In particular, the traveling wave propagation speed of the lines... The wave velocity is subject to slow changes due to various environmental and physical factors such as temperature, humidity, line icing, and line aging. If the wave velocity parameters in the model deviate from the actual wave velocity of the line, it will directly affect the forward propagation operator. Inaccurate predictions can reduce the final positioning accuracy. To address this technical problem, this embodiment proposes a feedback adjustment mechanism: active detection and medium model calibration.

[0149] The core idea of ​​this mechanism is to no longer passively wait for real faults (network disturbance singularities) to occur, but instead for the system to actively and periodically inspect the waveguide network medium. By analyzing the inspection results, the medium model is corrected and calibrated in reverse, enabling it to dynamically and in real-time approximate the real state of the physical network. The introduction of this mechanism upgrades the entire positioning system from a static open-loop system to a closed-loop intelligent system with adaptive and self-learning capabilities.

[0150] The implementation of this mechanism can be performed as a standalone method, such as... Figure 2 As shown, the main steps include:

[0151] In the distribution network corresponding to the S610 periodic guided wave network dielectric model, a calibration pulse signal with known characteristics is injected.

[0152] S620. The calibration transient interference spectrum triggered by the calibration pulse signal is synchronously recorded by multiple synchronous sampling array nodes deployed in the power distribution network.

[0153] S630. Extract the calibration spectrum feature vector from the calibration transient interferometric spectrum, and obtain the model error of the waveguide network medium model based on the difference between the calibration spectrum feature vector and a theoretical calibration spectrum feature vector predicted by the waveguide network medium model.

[0154] S640. Based on the model error, update at least one model parameter in the waveguide network medium model.

[0155] In one specific embodiment, the calibration pulse signal in step S610 is generated by a dedicated calibration pulse injection unit. This unit can be integrated into a critical synchronous sampling array node (e.g., a node at the substation busbar) or installed as a standalone device. This unit generates a low-energy pulse signal with well-defined waveform characteristics and high repeatability. The low energy ensures that the injection will not interfere with the normal operation of the power grid. The well-defined characteristics, for example, could be a square wave pulse with an extremely fast rise time and a defined pulse width. The system automatically triggers the unit at a preset period (e.g., every 6 hours) to inject this calibration pulse into the line it is connected to. The timing and location of the injection (denoted as...) Since the source is precisely known, this pulse signal becomes an ideal "known source".

[0156] In step S620, once the calibration pulse is injected, it propagates outwards along the distribution network topology like a real fault traveling wave. All synchronous sampling array nodes deployed in the network are triggered by this pulse signal (or forced to record within a specific time window by instructions from the central processing unit), and synchronously record the response waveforms they receive. These recorded waveforms, induced by the known calibration signal, are called the calibration transient interferogram.

[0157] In step S630, the system performs the same feature extraction process as in S200 on all acquired calibration transient interferometric spectra, obtaining a set of actual observed calibration spectrum feature vectors. At the same time, the system calls the forward propagation operator defined in S300. But this time, the input is the known location of the calibration pulse source. Through operators Through calculation, the system obtained a set of theoretically predicted calibration spectrum eigenvectors. This theoretical prediction is based on the current waveguide network dielectric model. (Including current wave speed parameters) The theoretical predictions are calculated using the model parameters (e.g., ...). Ideally, if the model were perfectly accurate, the theoretical predictions should exactly match the actual observations. However, due to variations in model parameters (e.g., ...), the actual predictions may differ from the actual observations. There is a bias, and a difference will appear between the two sets of vectors. The system quantifies the model error of the current model by calculating the difference between these two sets of vectors. Model error can be defined as the sum of the distances between the theoretical and actual vectors at all nodes: .

[0158] In step S640, the system utilizes the model error calculated in the previous step. This drives the updating of model parameters. The goal of this step is to adjust the model parameters (mainly the wave velocity of each line segment). This allows the adjusted new model to minimize model error. This is a typical optimization problem. In a preferred embodiment, a gradient-based optimization algorithm, such as gradient descent, can be used. The system will calculate the model error. For each parameter to be optimized gradient This gradient indicates the parameters that, in order to minimize the error most quickly. The direction that should be adjusted. Then, update the parameters according to the following rules:

[0159]

[0160] in, It is a learning rate hyperparameter that controls the step size of each update. By executing this update rule once or multiple times, the system can automatically and finely fine-tune the wave velocity parameters in the model, making the entire medium model increasingly approximate the true state of the physical network.

[0161] For example, suppose the system performs a routine calibration at 3 a.m.

[0162] The calibration unit located in the substation injected a standard square wave pulse into feeder No. 1.

[0163] All 10 sampling nodes in the network recorded the response waveforms and extracted their respective observation calibration spectrum feature vectors. .

[0164] The system uses the current model (assuming the wave velocity parameters of feeder 1). for The forward propagation operator (m / s) is used to calculate the theoretically predicted calibration spectrum eigenvector. .

[0165] Calculated model error The value was 12.4, which is higher than the warning threshold.

[0166] The system starts the optimization program. The error is calculated. Compared to The gradient is positive, which means that in the current model... The value is too low (leading to a later predicted arrival time, which in turn causes a bias in the feature vector).

[0167] According to the gradient descent update rule, the system will The value increased slightly and was updated to... m / s.

[0168] Recalculate using the updated model It was found that it dropped to 1.5, which met the accuracy requirements.

[0169] The system will assign new wave velocity values. Saved to the waveguide network medium model database for use in the next real fault location.

[0170] Through such an active, closed-loop calibration mechanism, the fault location method described in this application can overcome the challenges brought about by environmental changes and model uncertainties, and always maintain the high-precision model foundation required by its core algorithm, thereby achieving stable and reliable positioning performance under various actual working conditions.

[0171] Example 3

[0172] This embodiment provides a physical or logical structure for a power distribution network fault location terminal to implement the aforementioned method. The terminal can be a server deployed in a central processing unit or an edge device integrated with computing capabilities. Its core functions are implemented by a series of cooperating modules.

[0173] In one embodiment, such as Figure 3 As shown, the terminal includes: a data acquisition module, a feature extraction module, a reconstruction module, and a decoding module.

[0174] The acquisition module is responsible for communicating with multiple synchronous sampling array nodes deployed on-site, receiving transient interferometric spectrum data uploaded by them, managing the data reception protocol, parsing data packets, and aligning and buffering data from different nodes based on precise timestamps. This module ensures that the data input into subsequent processing flows is synchronous, complete, and error-free.

[0175] The feature extraction module receives raw transient interferometric spectrum data from the acquisition module. Internally, it integrates the complete processing pipeline described in S200. It first performs signal normalization preprocessing, including baseline correction and amplitude normalization. Subsequently, it invokes a series of parallel digital signal processing algorithms to calculate various time-domain morphological features, frequency-domain structural features, and time-frequency joint features of the waveform. In particular, it includes a high-efficiency wavelet transform engine for generating a time-spectrum graph and extracting graphical features from it. Finally, the module integrates all calculated feature values ​​into a standard-format spectral feature vector and outputs it to the reconstruction module.

[0176] The reconstruction module receives a set of spectral feature vectors from the feature extraction module, consisting of multiple observation nodes. Internally, this module stores a database of the entire distribution network's guided wave network medium model, including network topology, line parameters, and information on all discretized candidate singularities. Upon receiving a set of observation vectors, the module initiates a large-scale parallel computation task. For each candidate singularity, it invokes an internally implemented forward propagation operator to predict the theoretical spectral feature vector, and then calculates the holographic coherence index between the theoretical vector and the actual observation vector. This computation process can be highly parallelized, for example, executed on a multi-core CPU or GPU, to complete the traversal of all candidate singularities in a short time. The final output of this module is a holographic data volume of singularities containing all candidate singularities and their coherence indices.

[0177] The decoding module receives the singularity holographic data volume from the reconstruction module. This module first executes an efficient global maximum search algorithm to determine the fault location. Next, it performs local field analysis around the maximum point, calculating metrics such as peak sharpness to assess the confidence level of the location results. Optionally, this module may also include an expert system linked to a fault knowledge base to make a preliminary judgment on the fault type by analyzing residuals. Finally, this module integrates all analysis results into a human-readable, information-rich fault location report, which is presented to operations personnel through a user interface (such as a web interface or mobile application push notification).

[0178] Furthermore, for the system implementing Embodiment 2, the terminal also includes a model calibration module. This module is responsible for periodically instructing the calibration pulse injection unit to operate, coordinating all the aforementioned modules to complete a full model calibration process, and ultimately updating the media model database used by the reconstruction module. These modules work together to form an automated and intelligent fault location system from data acquisition to final decision-making.

[0179] Example 4

[0180] This embodiment provides a computer-readable storage medium, such as a non-volatile memory (e.g., hard disk, solid-state drive, flash memory), or computer memory. The storage medium stores computer program instructions. When these instructions are loaded and executed by one or more processors (e.g., a server's CPU, an embedded system's DSP, or an FPGA), the processor can implement all or part of the steps of the "Distribution Network Fault Location Method Combining Multi-Terminal Traveling Wave Sampling" described in Embodiment 1 or Embodiment 2.

[0181] Specifically, these program instructions are organized into multiple functional modules. A data reception and synchronization module's instruction set is used to execute data acquisition and management in S100. A signal processing and feature extraction module's instruction set, when executed, implements all algorithms in S200, transforming the original waveform into feature vectors. An iterative reconstruction calculation module's instruction set contains code implementing the core algorithms of S300, including access to the medium model, implementation of the forward propagation operator, and parallel calculation of the holographic coherence index. A decoding and report generation module's instruction set is used to implement localization, confidence assessment, and result presentation in S400. If the functionality of Embodiment 2 is included, there will also be a model adaptive calibration module's instruction set, used to execute the closed-loop optimization process from S610 to S640. By embedding the core method of this invention as a software program on a storage medium, it can be easily deployed, upgraded, and maintained on various general-purpose or dedicated computing hardware.

[0182] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another apparatus, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0183] The units described as separate components may or may not be physically separate. A component shown as a unit can be one physical unit or multiple physical units; that is, it can be located in one place or distributed in multiple different places. Depending on actual needs, some or all of the units can be selected to achieve the purpose of this embodiment.

[0184] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware.

[0185] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for fault location in a distribution network combining multi-terminal traveling wave sampling, characterized in that, include: By deploying multiple synchronous sampling array nodes in the distribution network, transient interference spectra excited by network disturbance singularities are collected synchronously. The multiple transient interferometric spectra are normalized, and a spectral feature vector is extracted from each transient interferometric spectrum. The normalization process includes: for each transient interferometric spectrum, calculating the average value of its pre-trigger recording portion and subtracting the average value from the entire data frame to complete baseline correction; and dividing the value of each sampling point in the data frame by the maximum absolute value of all sampling points in the data frame to complete amplitude normalization; and extracting a spectral feature vector from each normalized transient interferometric spectrum. In a pre-defined waveguide network medium model, for each candidate singularity in the model, a theoretical spectral feature vector is obtained for each synchronous sampling array node through a forward propagation operator. For each synchronous sampling array node, the dissimilarity between the spectral feature vector actually observed by the node and the corresponding theoretical spectral feature vector is calculated. The dissimilarity of all synchronous sampling array nodes is weighted and summed to obtain the total mismatch. The total mismatch is then converted into the holographic coherence index of the candidate singularity through an inverse function. All candidate singularities are traversed to generate singularity holographic data volumes. Decode the singularity holographic data volume to determine the location of the network disturbance singularity in the power distribution network; The singularity holographic data volume is composed of all the candidate singularities and their corresponding holographic coherence indices, and is a scalar field representing the distribution of the global fault probability on the waveguide network medium model topology.

2. The method according to claim 1, characterized in that, The method of synchronously acquiring transient interference spectra excited by network disturbance singularities through multiple synchronous sampling array nodes deployed in the distribution network includes: The multiple synchronous sampling array nodes achieve time synchronization based on a unified high-precision clock source; After the network disturbance singularity occurs, the multiple synchronous sampling array nodes are activated by a preset trigger logic, and each captures and stores a data frame of a preset length as the transient interference spectrum.

3. The method according to claim 1, characterized in that, The step of extracting a spectral feature vector from each of the transient interferometric spectra includes: The transient interference spectrum is subjected to a time-frequency joint transformation to generate a two-dimensional time-frequency spectrum. At least one time-frequency joint feature is extracted from the two-dimensional time-spectrum graph, and the time-frequency joint feature is used as a component of the spectral feature vector.

4. The method according to claim 1, characterized in that, The process of obtaining the holographic coherence index of the candidate singularity includes: Using a forward propagation operator, based on the position of the candidate singularity in the waveguide network medium model, the theoretical spectral eigenvector that should be observed at each of the synchronous sampling array nodes is predicted; The holographic coherence index of the candidate singularity is obtained by combining the vector distance or similarity between the theoretical spectral feature vector and the spectral feature vector.

5. The method according to claim 1, characterized in that, Decoding the singularity holographic data volume to determine the location of the network disturbance singularity in the power distribution network includes: In the scalar field represented by the singularity holographic data volume, search for candidate singularities with the global maximum holographic coherence index; The position of the candidate singularity with the global maximum holographic coherence index is determined as the position of the network perturbation singularity.

6. A method for locating faults in a distribution network, comprising the method for locating faults in a distribution network combining multi-terminal traveling wave sampling as described in claim 1, characterized in that, include: A calibration pulse signal with known characteristics is periodically injected into a distribution network corresponding to a waveguide network dielectric model. Multiple synchronous sampling array nodes deployed in the power distribution network synchronously record the calibration transient interference spectrum induced by the calibration pulse signal; The calibration spectrum feature vector is extracted from the calibration transient interferometric spectrum, and the model error of the waveguide network medium model is obtained based on the difference between the calibration spectrum feature vector and a theoretical calibration spectrum feature vector predicted by the waveguide network medium model. Based on the model error, at least one model parameter in the waveguide network medium model is updated.

7. The method according to claim 6, characterized in that, The at least one model parameter includes the traveling wave velocity of the distribution network line.

8. The method according to claim 6, characterized in that, The step of updating at least one model parameter in the waveguide network dielectric model based on the model error includes: An optimization algorithm is used to adjust the at least one model parameter to minimize the difference.

9. A distribution network fault location terminal, used to implement the distribution network fault location method combining multi-terminal traveling wave sampling as described in claim 1, characterized in that, include: A data acquisition module is used to synchronously acquire transient interferometric spectra excited by network disturbance singularities through multiple synchronous sampling array nodes deployed in the distribution network. A feature extraction module is used to normalize the plurality of transient interferometric spectra and extract a spectral feature vector from each of the transient interferometric spectra. The normalization process includes: for each of the transient interferometric spectra, calculating the average value of its pre-trigger recording portion and subtracting the average value from the entire data frame to complete baseline correction; and dividing the value of each sampling point in the data frame by the maximum value of the absolute values ​​of all sampling points in the data frame to complete amplitude normalization; and extracting a spectral feature vector from each of the normalized transient interferometric spectra. A reconstruction module is used to obtain the corresponding theoretical spectral feature vector for each candidate singularity in a preset waveguide network medium model, and obtain the holographic coherence index of the candidate based on the theoretical spectral feature vector and the spectral feature vector, so as to generate a singularity holographic data volume. A decoding module is used to decode the singularity holographic data volume to determine the location of the network disturbance singularity in the power distribution network.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the power distribution network fault location method combined with multi-terminal traveling wave sampling as described in any one of claims 1-5 or the power distribution network fault location method as described in any one of claims 6-8.

Images