Fault arc detection method, apparatus, and device

Abstract

This invention provides a method, apparatus, and device for detecting fault arcs, relating to the field of fault arc measurement technology in power distribution lines. The method includes: acquiring multiple types of disturbance signals from a target power distribution line; determining a disturbance analysis time window; and extracting basic ranging-related features based on the multiple types of disturbance signals within the time window; generating output features for each branch of the fault arc propagation through a preset neural network model corresponding to different physical branches of the fault arc propagation, and fusing the output features of each branch with the basic ranging-related features to obtain an enhanced ranging feature vector; inputting the enhanced ranging feature vector into a preset ranging network to output the initial distance, residual sensitive features, and state reliability of the fault arc; correcting the initial distance based on the residual sensitive features to obtain a corrected distance; and determining the fault arc detection result based on the corrected distance and state reliability. This invention can provide accurate and reliable information related to the location of fault arcs.

Description

Technical Field

[0001] This invention relates to the field of power distribution line fault arc measurement technology, and in particular to a fault arc detection method, device and equipment. Background Technology

[0002] Electric arc faults are a common and dangerous type of fault in low-voltage power distribution systems and electrical equipment. They are typically caused by factors such as aging insulation, loose terminals, conductor damage, poor contact, or air breakdown in humid environments. Electric arc faults are characterized by their high randomness, unstable duration, significant waveform distortion, abundant high-frequency disturbances, and susceptibility to electrical fires.

[0003] Most existing arc fault measurement technologies focus solely on identifying whether an arc fault has occurred. For example, some solutions integrate arc fault detection into electrical equipment such as electricity meters and circuit breakers. By collecting voltage and current data and determining whether high-frequency harmonic parameters exceed thresholds, they identify arc faults. However, these solutions only provide basic fault alarm functions. In actual low-voltage power systems, arc faults can occur at multiple locations, including branch ends, socket interfaces, load connection points, and the middle of the line. Maintenance personnel cannot quickly and accurately locate the fault point relying solely on equipment alarm information. Furthermore, parameters such as line length, conductor cross-sectional area, branch topology, load operating status, power supply equivalent impedance, and ambient temperature all alter the propagation characteristics of arc disturbances. Relying on a single fault detection method cannot provide a stable and accurate reference for fault location.

[0004] In addition, existing fault arc location schemes usually rely on a single current signal, a single voltage signal, or high-frequency characteristics of a fixed frequency band for location estimation. Under conditions of complex low-voltage branch topology and multiple loads operating simultaneously, interference factors such as load start-up and shutdown, switch opening and closing, power electronic equipment operation, and sampling noise can easily cause signal characteristic distortion, resulting in poor accuracy and stability of fault location. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method, apparatus, and device for detecting fault arcs, so as to provide accurate and reliable information related to the location of fault arcs.

[0006] In a first aspect, embodiments of the present invention provide a method for detecting fault arcs, comprising: Acquire various disturbance signals from the target power distribution line; Based on multiple types of disturbance signals, a disturbance analysis time window is determined, and basic ranging-related features are extracted based on the multiple types of disturbance signals within the disturbance analysis time window. Based on multiple types of disturbance signals, the output features of each branch are generated through a preset neural network model corresponding to different fault arc propagation physical branches. The output features of each branch are then fused with the basic ranging-related features to obtain an enhanced ranging feature vector. The preset neural network model corresponding to each branch has different time selection operators and time mixing blocks. The enhanced ranging feature vector is input into the preset ranging network, and the initial distance of the fault arc, residual sensitive features, and state confidence are output. The initial distance is corrected based on the residual sensitivity characteristics to obtain the corrected distance. Based on the corrected distance and the state confidence level, the fault arc detection result is determined.

[0007] In one possible implementation, the disturbance analysis time window is determined based on multiple types of disturbance signals, including: Determine the combined disturbance intensity of multiple disturbance signals; If the overall disturbance intensity is greater than the preset intensity threshold at any time, then monitor the duration of the overall disturbance intensity being greater than the preset intensity threshold, and calculate the high-frequency energy ratio and channel consistency index of multiple disturbance signals. If at least two of the duration, high-frequency energy ratio, and channel consistency indicators meet the corresponding fault arc judgment conditions, then that moment is determined as the disturbance arrival reference moment, and the disturbance analysis time window is determined based on the disturbance arrival reference moment.

[0008] In one possible implementation, based on multiple types of disturbance signals, a preset neural network model corresponding to different fault arc propagation physical branches is used to generate output features for each branch. These branch output features are then fused with basic ranging-related features to obtain an enhanced ranging feature vector, including: Multi-channel features are obtained by embedding multiple types of perturbation signals using a shared input. The multi-channel features are input into the preset neural network model corresponding to each physical branch of the fault arc propagation to obtain the time-series features corresponding to each branch. Construct a preset physical feature vector corresponding to each physical branch of fault arc propagation, and fuse the preset physical feature vectors corresponding to each branch with the time series features to obtain the output features of each branch; The output features of each branch are fused to obtain the enhanced output features; The enhanced output features are fused with the basic ranging-related features to obtain the enhanced ranging feature vector.

[0009] In one possible implementation, the ranging network includes: a ranging state path, an interference state path, and an interference suppression gating; The enhanced ranging feature vector is input into a pre-defined ranging network, which outputs the initial distance of the fault arc, residual sensitive features, and state confidence level, including: The enhanced ranging feature vector is input into the ranging state path to obtain the distance state features; The enhanced ranging feature vector, along with the pre-acquired power metering parameters and temperature drift suppression parameters, are input into the interference state path to obtain the interference state characteristics. The distance state features and interference state features are input into the interference suppression gating so that the distance state features are corrected by the interference state features; The corrected distance state characteristics are input into three preset output heads to obtain the initial distance, residual sensitivity characteristics, and state reliability of the fault arc output by the three output heads.

[0010] In one possible implementation, the initial distance is corrected based on residual sensitivity features to obtain the corrected distance, including: Based on the residual sensitivity characteristics and the pre-acquired power metering parameters and temperature drift suppression parameters, determine the distance residual correction amount corresponding to the line parameters, load status, calibration compensation and temperature drift suppression; The initial distance is corrected based on the distance residual correction amount corresponding to line parameters, load status, calibration compensation, and temperature drift suppression, to obtain the corrected distance.

[0011] In one possible implementation, the fault arc detection result is determined based on the correction distance and state confidence level, including: The distance reliability is determined based on the state reliability, as well as the pre-acquired channel disturbance synchronization degree, disturbance signal-to-noise ratio, line parameter integrity and load state stability indicators. Based on the distance reliability, as well as the pre-obtained sampling period error, line calibration error, state uncertainty index, and minimum error lower limit, the final error range is determined; Based on the correction distance, distance reliability, and final error range, fault arc detection results are generated.

[0012] In one possible implementation, the various types of disturbance signals include: current disturbance signals, voltage disturbance signals, and near-field electromagnetic disturbance signals.

[0013] In one possible implementation, the different physical branches of fault arc propagation include: The first-arrival mutation branch is used to reflect the mutation information generated when the fault arc first arrives; The propagation attenuation branch is used to reflect the attenuation information of the fault arc during its propagation along the line; Topological reflection branch is used to reflect the reflection changes caused by abrupt changes in the equivalent impedance of line nodes; The slow-change branch of the load is used to reflect changes in load status parameters.

[0014] Secondly, embodiments of the present invention provide a fault arc detection device, comprising: A multi-channel sensing module is used to acquire various disturbance signals of the target power distribution line; The disturbance analysis module is used to determine the disturbance analysis time window based on multiple types of disturbance signals, and to extract basic ranging-related features based on the multiple types of disturbance signals within the disturbance analysis time window; The feature enhancement module is used to generate output features of each branch based on multiple types of disturbance signals through a preset neural network model corresponding to different fault arc propagation physical branches, and to fuse the output features of each branch with the basic ranging-related features to obtain an enhanced ranging feature vector; wherein, the preset neural network model corresponding to each branch has different time selection operators and time mixing blocks. The ranging module is used to input the enhanced ranging feature vector into the preset ranging network and output the initial distance of the fault arc, residual sensitive features and state confidence. The correction module is used to correct the initial distance based on the residual sensitivity characteristics to obtain the corrected distance. Based on the corrected distance and the state confidence level, the fault arc detection result is determined.

[0015] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.

[0016] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows: This invention first acquires multiple disturbance signals from a power distribution line, defines a disturbance analysis time window adapted to the disturbance waveform, and extracts basic ranging features based on the multiple disturbance signals within the window. Then, it employs a multi-branch neural network with differentiated time selection operators and time mixing blocks, corresponding to different fault arc propagation mechanisms, to generate dedicated output features for each physical branch. These branch output features are then fused with the basic ranging features to construct an enhanced ranging feature vector. This enhanced ranging feature vector is input into the ranging network, which simultaneously outputs the initial distance of the fault arc, residual sensitive features, and state reliability. The initial distance is then dynamically corrected based on the residual sensitive features to obtain the corrected distance and output a complete fault detection result. This invention, relying on a multi-physical branch feature enhancement mechanism that conforms to the arc propagation mechanism, effectively improves the physical targeting of ranging feature extraction. Simultaneously, it utilizes residual sensitive features to distinguish effective ranging information from various types of interference, significantly improving ranging stability under complex line conditions. Compared to traditional detection schemes that can only determine whether a fault arc has occurred, this invention can accurately determine the location of the fault arc, providing reliable data support for maintenance personnel to quickly delineate fault sections and accurately locate arc fault points, thereby improving the efficiency of line fault repair. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the implementation of the fault arc detection method provided in this embodiment of the invention; Figure 2 This is a schematic diagram of the physical branch of fault arc propagation provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of the dual-channel ranging network provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the fault arc detection device provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0019] To illustrate the technical solution described in this invention, specific embodiments are described below.

[0020] Existing technologies integrate arc fault detection into electricity meters or circuit breakers, focusing on determining whether an arc fault has occurred based on voltage and current sampling results and whether high-frequency harmonics meet thresholds. While these solutions can provide fault alarms, they primarily output fault events and struggle to provide further information such as the distance between the arc fault point and the measuring end, the error range, and the reliability of the distance.

[0021] This invention, through its embodiments, acquires the current disturbance, voltage disturbance, and near-field electromagnetic disturbance generated by the propagation of a fault arc along the tested power circuit, and extracts basic ranging features such as arrival time difference, propagation attenuation, phase offset, disturbance energy, and rising edge slope. Furthermore, it decomposes the disturbance propagation process into different physical scales, such as first-arrival abrupt change, propagation attenuation, topological reflection, and slow load variation, through a propagation physical scale branch, and performs inter-scale hybrid enhancement in the feature enhancement structure. Subsequently, it outputs the initial distance of the fault arc, residual sensitive features, and state reliability through a dual-path ranging network composed of the ranging state path and the disturbance state path. Reliable residual correction is then performed by combining line parameters, power metering parameters, calibration compensation parameters, and temperature drift suppression parameters, ultimately outputting information such as the fault arc distance, error range, and distance reliability.

[0022] See Figure 2The flowchart illustrating the implementation of the fault arc detection method provided in this embodiment of the invention is shown below: Step S101: Obtain various disturbance signals of the target power distribution line.

[0023] Here, various types of disturbance signals can include: current disturbance signals, voltage disturbance signals, and near-field electromagnetic disturbance signals. This embodiment sets up a current disturbance sensing channel to collect current surges, high-frequency pulses, and waveform distortion signals that propagate along the line to the measurement end after a fault arc occurs; a voltage disturbance sensing channel to collect instantaneous voltage drops, spikes, and gap disturbances at the line terminals; and a near-field electromagnetic disturbance sensing channel to collect electromagnetic disturbance signals generated during the fault arc discharge process and coupled to the vicinity of the measurement end.

[0024] The three types of disturbance signals can be represented as: in, This indicates a current disturbance signal. This indicates a voltage disturbance signal. This represents the near-field electromagnetic disturbance signal, where t is the sampling instant, representing the continuous time variable of the disturbance signal in the power distribution line.

[0025] Let the sampling frequency be The sampling period is: Then the first The time corresponding to each sampling point is: After discretizing the three types of disturbance signals, a multi-channel disturbance sampling sequence can be obtained: To facilitate subsequent extraction of the disturbance arrival time, baseline correction and normalization processing of the sampled signal are also required, using any channel. For example, the normalized perturbation signal can be expressed as: Where n is the discrete sampling point number, Indicates the first The average value of each channel under normal operating conditions. Indicates standard deviation, To prevent extremely small constants with a denominator of zero, This represents the original sampled signal at point n in channel c. This represents the normalized perturbation signal at point n of channel c.

[0026] In addition, this embodiment also obtains the basic parameters of the target power distribution line in advance for subsequent distance calculation, topology reflection identification and residual correction.

[0027] The basic parameters of the line include the total length of the line, conductor type, conductor cross-sectional area, resistance per unit length of the line, reactance per unit length of the line, equivalent impedance of the line, branch topology, load connection location of each branch, reference propagation speed, reference calibration temperature, and historical calibration coefficients.

[0028] For example, the set of basic parameters for a line is represented as: in, Indicates the total length of the line. Represents the cross-sectional area of ​​the conductor. This represents the resistance per unit length of the line. Represents the reactance per unit length of the line. This represents the equivalent impedance of the line. Represents branch topology information, Indicates the reference propagation speed of the disturbance. Indicates the historical calibration coefficient. Indicates the reference calibration temperature.

[0029] Step S102: Determine the disturbance analysis time window based on the various disturbance signals, and extract the basic ranging correlation features based on the various disturbance signals within the disturbance analysis time window.

[0030] In one possible implementation, the step may include: determining the overall disturbance intensity of multiple disturbance signals; if the overall disturbance intensity is greater than a preset intensity threshold at any time, then monitoring the duration for which the overall disturbance intensity is greater than the preset intensity threshold, and calculating the high-frequency energy ratio and channel consistency index of the multiple disturbance signals; if at least two of the duration, high-frequency energy ratio, and channel consistency index satisfy the corresponding fault arc judgment condition, then determining the time as the disturbance arrival reference time, and determining the disturbance analysis time window based on the disturbance arrival reference time.

[0031] Specifically, for the c-th disturbance channel, the short-time disturbance energy is defined as: Where w represents the preset short-time energy calculation window length, n is the discrete sampling point number, and i is the local sampling point index within the short-time energy calculation window, used to calculate the short-time disturbance energy point by point.

[0032] The combined intensity of the multi-channel disturbance can be expressed as: in, I This indicates the current disturbance signal channel. U This indicates the voltage disturbance signal channel. MThis indicates the near-field electromagnetic disturbance signal channel; 、 and Let represent the combined weights of the current disturbance channel, voltage disturbance channel, and near-field electromagnetic disturbance channel, respectively, satisfying: When the overall disturbance intensity first exceeds a preset intensity threshold, and at least two of the disturbance duration, high-frequency energy percentage, and channel consistency indicators meet the fault arc disturbance criterion, it is determined as the starting point of a fault arc disturbance event. The disturbance event window represents the starting point of the fault arc disturbance event. Once determined, a sampling interval is used to search for the arrival points of disturbances in each channel. This is to avoid the first arrival point of a single channel arriving slightly earlier than the last. If it is missed, the event window can be accessed. A certain number of sampling points are retained before and after, defined as: in, The window displays the disturbance event; and point This indicates the start and end sampling points of the event window; Indicates the number of sampling points reserved before the event's starting point; Indicates the number of sampling points retained after the event's origin; This indicates the total number of sampling points in the current sampling sequence.

[0033] In the disturbance event window Within this embodiment, the basic ranging-related features are extracted as follows: The arrival points and arrival times of disturbances for each channel are extracted separately. The arrival points for each channel can be represented as follows: The arrival times of each channel can be expressed as: in, Indicates the first Disturbances from each channel reach the sampling point; Indicates the first The arrival time of disturbances in each channel; Indicates the first The first channel in the Normalized perturbation signal at each sampling point; Indicates the first Short-term disturbance energy of each channel; Indicates the first The amplitude perturbation threshold for each channel; surface Show the first Short-time energy threshold of each channel; Indicates the sampling period.

[0034] This allows us to determine the arrival times of current disturbances, voltage disturbances, and near-field electromagnetic disturbances: 。

[0035] After obtaining the arrival times of the three types of disturbance signals, distance-related propagation features are further extracted, including arrival time difference, disturbance peak value, disturbance energy, propagation attenuation, phase shift, and rising edge slope.

[0036] The time difference of arrival between different disturbance channels can be expressed as: The peak value and energy of the disturbance in each channel can be expressed as: in, 、 and These represent the arrival time differences between the current disturbance channel and the voltage disturbance channel, the current disturbance channel and the near-field electromagnetic disturbance channel, and the voltage disturbance channel and the near-field electromagnetic disturbance channel, respectively. 、 and These represent the arrival times of current disturbance, voltage disturbance, and near-field electromagnetic disturbance, respectively. surface Show the first The peak value of the disturbance in each disturbance channel within the disturbance event window; Indicates the first Each disturbance channel in the disturbance event window Internal disturbance energy; Indicates the first The first channel in the Normalized perturbation signal at each sampling point; The channel types are indicated as current disturbance channel, voltage disturbance channel, and near-field electromagnetic disturbance channel, respectively.

[0037] Propagation attenuation includes the amplitude attenuation between different disturbance channels. δA and energy decay δE , can be represented as: in, To prevent extremely small constants with a denominator of zero; / / Peak disturbance values ​​within the current / voltage / electromagnetic channel window; / / Total disturbance energy of each channel window; / : Current-voltage, current-electromagnetic channel amplitude attenuation coefficient; / : Current-voltage, current-electromagnetic channel energy attenuation coefficient.

[0038] The slope of the rising edge of the disturbance can be expressed as: in, This represents the sampling point corresponding to the peak value of the c-th channel.

[0039] For current and voltage disturbances, the target frequency phase within the event window can be calculated first, and then the phase offset can be obtained: in, The fundamental phase of the channel c disturbance signal. j The imaginary unit of complex numbers. This is the power frequency reference frequency for power distribution lines. Ts The sampling period.

[0040] Based on the above characteristics, a basic ranging feature vector is constructed to characterize the distance information carried by the fault arc disturbance as it propagates along the line to the measuring end: Step S103: Based on multiple types of disturbance signals, the output features of each branch are generated through the preset neural network model corresponding to different fault arc propagation physical branches, and the output features of each branch are fused with the basic ranging-related features to obtain the enhanced ranging feature vector; wherein, the preset neural network model corresponding to each branch has different time selection operators and time mixing blocks.

[0041] Let the sequence of multiple disturbance signals be: The multi-channel normalized perturbation sampling sequence within the perturbation event window can be represented as: in, 、 and These represent the current disturbance channel, voltage disturbance channel, and near-field electromagnetic disturbance channel respectively in the [missing information - likely a number]th [missing information - likely a number]. Normalized perturbation signal at each sampling point.

[0042] This embodiment inputs the multi-channel disturbance sampling sequence X into the TimeMixer feature enhancement architecture. Instead of directly using TimeMixer as a standard multi-scale time series prediction model, this architecture divides the fault arc disturbance propagation process into four branches based on the physical meaning of ranging: the first-arrival abrupt change branch, the propagation attenuation branch, the topological reflection branch, and the load slow variation branch. The first-arrival abrupt change branch characterizes the initial disturbance propagating along the line to the measurement end at the moment the fault arc occurs, including the first arrival point, rising edge abrupt change, peak initiation point, and transient energy abrupt change. The propagation attenuation branch characterizes the attenuation characteristics of amplitude, energy, and peak value as the disturbance propagates along the line. The topological reflection branch characterizes reflected disturbances, secondary peaks, trailing waveforms, and phase disturbances caused by abrupt changes in equivalent impedance at branch nodes, load connection points, line ends, or the power supply side. The load slow variation branch characterizes load periodic fluctuations, power factor changes, equivalent impedance changes, and steady-state distortion trends, used to distinguish between load disturbances and distance disturbances in subsequent ranging processes.

[0043] by Figure 2 For reference, this step may include: Step S1031: Shared input embedding is performed on multiple types of disturbance signals to obtain multi-channel features.

[0044] To enable unified processing of the four physical branches mentioned above within the neural network, TimeMixer first performs shared input embedding on the multi-channel perturbation sampling sequences: in, This represents the multi-channel temporal features after input embedding; and These represent the weights and biases of the input embedding layer, respectively. This represents a non-linear activation function.

[0045] Step S1032: Input the multi-channel features into the preset neural network model corresponding to each fault arc propagation physical branch to obtain the time-series features corresponding to each branch.

[0046] To reflect different propagation physical scales, TimeMixer uses distance-related scale selection to set different time selection operators and time mixing blocks for the four branches. Let... Let these represent the first-reach mutation branch, the propagation attenuation branch, the topological reflection branch, and the load-slow variation branch, respectively. Then the... The time-mixing block of each branch can be abstractly represented as: in, For input, Indicates the first Time-mixed blocks of branches; Indicates nonlinear activation function; This indicates the presence of a time-mixed nucleus or receptive field. 、 downsampling scale Time-mixing operators; surface Normalization is shown. The first mutation branch uses a smaller receptive field and a shorter time window. The propagation attenuation branch employs a medium receptive field and a full event window. The topological reflection branch uses a delay window after the first arrival. With a large receptive field, the slow-varying load branch uses long-term scale features after downsampling or low-frequency smoothing.

[0047] The neural network inputs corresponding to the four branches can be represented as: in, This indicates that timing features are captured according to the specified window. This indicates downsampling, moving average, or low-frequency smoothing. 、 、 and point Do not represent the timing input of the four branches.

[0048] After time mixing and pooling, the four branches yield the corresponding temporal representations: in, This indicates a pooling or sequence aggregation operation, used to aggregate information in the time dimension into branch-level features.

[0049] Step S1033: Construct the preset physical feature vectors corresponding to each physical branch of fault arc propagation, and fuse the preset physical feature vectors and time-series features corresponding to each branch to obtain the output features of each branch.

[0050] The temporal representations of each branch are fused with their corresponding physical feature vectors to obtain four branch outputs: in, 、 、 and These represent the first-arrival mutation characteristics, propagation attenuation characteristics, topological reflection characteristics, and load slow variation characteristics, respectively. 、 Each branch represents a preset physical feature vector; 、 、 and Each branch represents its output mapping network; This indicates feature splicing.

[0051] Step S1034: The output features of each branch are fused to obtain enhanced output features.

[0052] Here, the output features of each branch can be averaged and fused. Alternatively, in one possible implementation, the fusion weights of the four propagation physical branches can be determined based on multi-channel propagation consistency, perturbation signal-to-noise ratio, load stability, topological reflection reliability, and historical calibration error. First, construct the scale-selection input vector: in, This indicates consistency in multi-channel propagation. This represents the perturbation signal-to-noise ratio. surface Indicates load stability. Indicates the confidence level of topological reflection. This indicates historical calibration error.

[0053] A fusion score for the four propagation physics branches is generated using a distance-dependent scale selection function: in, The fusion function, if expressed in linear scale form, can be represented as: in, 、 、 and These represent the fusion scores for the first mutation branch, propagation attenuation branch, topological reflection branch, and load-slow variation branch, respectively. This represents the scale selection weight matrix; This represents the scale selection bias vector. These parameters can be obtained through training with historical experimental data, or they can be preset based on line parameters and calibration experiments.

[0054] The final fusion weights of the four branches are determined by the softmax function: Among them, the The softmax weights of each branch are: Therefore, the four fusion weights satisfy: In other words, the role of softmax is to normalize the scores of the four branches into non-negative weights, rather than directly specifying the weight values. When a branch is more reliable under the current disturbance state, its score is higher, and the fusion weight obtained after softmax is also increased accordingly; when a branch is more affected by noise, load fluctuations, or invalid reflections, its score decreases, and the corresponding fusion weight decreases.

[0055] Because the first-to-break mutation branch, propagation attenuation branch, topological reflection branch, and load-variable branch use different time windows, receptive fields, and auxiliary inputs, their output dimensions may differ and cannot be directly added. Therefore, before fusion, they are mapped to a unified dimension using a branch feature mapping matrix: in, 、 、 and These represent the feature mapping matrices or dimension alignment matrices of the four branches, respectively. 、 、 and surface This shows the branch features after mapping to a unified dimension. If the four branches already have the same dimension, this can also be done. 、 、 and Take it as a unit mapping.

[0056] After dimensional alignment is completed, the enhanced output features of TimeMixer, which propagates the physical scale, can be represented as: in, This represents the enhanced features of the TimeMixer output, which propagates at the physical scale.

[0057] Through the above steps, TimeMixer can enhance the contribution of different propagation components such as first arrival, attenuation, reflection, and slow load variation to ranging while preserving the basic physical ranging characteristics.

[0058] Step S1035: The enhanced output features are fused with the basic ranging-related features to obtain the enhanced ranging feature vector.

[0059] The enhanced ranging feature vector is represented as: in, Represents the basic ranging feature vector; This represents the enhanced ranging feature fusion mapping function; This represents the enhanced ranging feature vector input to the subsequent dual-path Mamba ranging state inversion module.

[0060] Step S104: Input the enhanced ranging feature vector into the preset ranging network and output the initial distance, residual sensitive features and state confidence of the fault arc.

[0061] This embodiment separates and models distance-related states and non-ranging interference states using a dual-path structure. The ranging state path selectively retains propagation states related to the distance to the fault arc, including first-arrival state, inter-channel arrival difference state, propagation attenuation state, phase drift state, and effective topology path state. The interference state path models and suppresses non-ranging states, including load cycle fluctuations, random noise, switching transients, non-fault electromagnetic interference, slow temperature drift, and invalid topology reflection states. The output of the interference state path is not directly used as the ranging result; instead, interference suppression gating is used to suppress and correct the output of the ranging state path.

[0062] by Figure 3 For reference, this step may include: Step S1041: Input the enhanced ranging feature vector into the ranging state path to obtain the distance state feature.

[0063] First, the augmented ranging feature vectors are converted into state embedding features that can be processed by Mamba through an input mapping layer: in, This represents the state features after input embedding; and point The weights and biases of the input mapping layer are not represented separately. This represents a non-linear activation function.

[0064] Ranging status path with As input, the focus is on extracting state information directly related to distance inversion, including first-arrival state, inter-channel arrival difference state, propagation attenuation state, phase drift state, and effective topology path state. Its input and output can be represented as: in, Indicates the input characteristics of the ranging state path; This represents the distance-related state characteristics output by the ranging state path; This represents the Mamba state-space network in the ranging state path.

[0065] Step S1042: Input the enhanced ranging feature vector and the pre-acquired power metering parameters and temperature drift suppression parameters into the interference state path to obtain the interference state features.

[0066] The disturbance state path is used to extract non-distance states. To enable this path to sense load and temperature states, energy metering parameters can be used. Temperature drift suppression parameters As an auxiliary input, it is fed into the perturbation state path along with the state embedding features. Its input can be represented as: in, Indicates the input characteristics of the interference state path; This indicates the non-ranging interference state characteristics of the interference state output path; Indicates the parameters for electricity metering; Indicates the temperature drift suppression parameter; surface Feature splicing; and surface Indicate the input mapping parameters of the interference path; This represents the Mamba state-space network in the interference state path.

[0067] Step S1043: Input the distance state features and interference state features into the interference suppression gate so as to correct the distance state features through the interference state features.

[0068] Within each Mamba path, the input features are sequentially normalized, projected, mixed locally in time, selectively state-space scanned, and projected back to obtain the corresponding state features. To suppress or correct the ranging state using interference states, an interference suppression gating unit is implemented. This gating unit generates dimension-wise gating coefficients based on the ranging state features and interference state features. in, Indicates interference suppression gating; Represents the Sigmoid function; and surface Display gate parameters. The range of values ​​is The larger the value, the stronger the influence of non-ranged interference on the corresponding state dimension.

[0069] The interference suppression gating method is used to suppress the ranging state dimension by dimension, which can be expressed as: in, This indicates the ranging state characteristics after interference suppression; This represents element-wise product. The formula indicates that when a certain state dimension is strongly disturbed, the contribution of the corresponding ranging state is reduced; when the disturbance is weak, more of the ranging state is preserved.

[0070] Subsequently, the suppressed ranging state features and the interference state features are input together into the state correction layer to obtain the corrected ranging state features: in, This indicates the corrected ranging status characteristics; and This represents the state correction mapping parameters. Through this structure, the interfering state is not used to replace the ranging state, but rather to suppress, filter, and correct the ranging state.

[0071] Step S1044: Input the corrected distance state characteristics into three preset output heads to obtain the initial distance, residual sensitive characteristics and state reliability of the fault arc output by the three output heads.

[0072] The dual-path Mamba outputs the initial distance of the fault arc, residual sensitivity characteristics, and state confidence level through three output heads. The initial distance output head can be represented as: The output header of the residual sensitive feature can be represented as: The state credibility output header can be represented as: in, surface Indicates the initial distance of the fault arc before correction; This indicates a residual sensitivity characteristic, used to characterize that the current initial distance error is more likely to originate from line parameters, load status, channel calibration error, or temperature drift. Indicates the reliability of dual-path Mamba states; 、 、 、 、 and This indicates the parameters of the corresponding output header.

[0073] In one possible implementation, to prevent the initial distance from exceeding the range of the measured line, a line length boundary constraint can be applied to the initial distance: in, Indicates the total length of the measured line. This represents the initial distance after the line length boundary constraint.

[0074] Through the aforementioned dual-path structure, Mamba is no longer used only as a regular sequence regression model, but is used for distance state selection and interference state suppression in fault arc ranging. This preserves distance-related states such as first arrival, arrival difference, attenuation, and phase drift, while weakening non-ranging states such as load background, random noise, non-fault transients, and slow temperature drift.

[0075] Step S105: Correct the initial distance according to the residual sensitivity characteristics to obtain the corrected distance. Based on the corrected distance and the state confidence level, determine the fault arc detection result.

[0076] Here, the initial distance after boundary constraints from the dual-path Mamba output can be used as a reference. 、 Residual Sensitive Features and state credibility By combining line parameters, power metering parameters, calibration compensation parameters, and temperature drift suppression parameters, reliable residual compensation is performed on the initial distance to obtain the corrected distance. This correction process does not re-detect the fault. Instead, based on the initial ranging results, it determines that the current ranging error is more likely to originate from line parameter deviations, load state changes, historical calibration errors, or temperature drift, and calculates the residual correction amount and dynamic correction weight accordingly.

[0077] To make the ranging results more suitable for actual maintenance use, this embodiment further calculates the distance reliability and error range. The final ranging results may include, but are not limited to: corrected distance, error range, distance reliability, distance interval, branch where the fault is located, fault occurrence time, and alarm signal.

[0078] This invention first acquires multiple disturbance signals from a power distribution line, defines a disturbance analysis time window adapted to the disturbance waveform, and extracts basic ranging features based on the multiple disturbance signals within the window. Then, it employs a multi-branch neural network with differentiated time selection operators and time mixing blocks, corresponding to different fault arc propagation mechanisms, to generate dedicated output features for each physical branch. These branch output features are then fused with the basic ranging features to construct an enhanced ranging feature vector. This enhanced ranging feature vector is input into the ranging network, which simultaneously outputs the initial distance of the fault arc, residual sensitive features, and state reliability. The initial distance is then dynamically corrected based on the residual sensitive features to obtain the corrected distance and output a complete fault detection result. This invention, relying on a multi-physical branch feature enhancement mechanism that conforms to the arc propagation mechanism, effectively improves the physical targeting of ranging feature extraction. Simultaneously, it utilizes residual sensitive features to distinguish effective ranging information from various types of interference, significantly improving ranging stability under complex line conditions. Compared to traditional detection schemes that can only determine whether a fault arc has occurred, this invention can accurately determine the location of the fault arc, providing reliable data support for maintenance personnel to quickly delineate fault sections and accurately locate arc fault points, thereby improving the efficiency of line fault repair.

[0079] In one possible implementation, step S1033, which constructs a preset physical feature vector corresponding to each physical branch of fault arc propagation, may include: (1) First mutation branch The first-arrival mutation branch is used to characterize the mutation information generated when the fault arc disturbance first arrives at the measurement end, including the arrival time of each channel, the arrival time difference between channels, the rising edge slope, the peak initiation point, and the disturbance peak value. Let the earliest arrival point among the three channels be: The first mutation branch uses a short time window near the first point of arrival: in, This indicates the first mutation analysis window. This indicates the number of sampling points reserved before the first arrival point. This indicates the number of sampling points retained after the first arrival point.

[0080] No. The rising edge slope characteristics of each channel can be expressed as: The physical eigenvector of the first mutation branch can be represented as: in, 、 and These represent the arrival times of the disturbances in the three channels, respectively. 、 and Indicates the time difference of arrival between channels; 、 and This indicates the rising edge slope characteristics of the three channels; 、 and This indicates the peak disturbance value of the three channels within the event window.

[0081] (2) Propagation attenuation branch The propagation attenuation branch is used to characterize the attenuation information of the amplitude, energy, and peak value of a disturbance as it propagates along the path. Each channel in the event window The internal disturbance energy is ,aisle With channel The amplitude attenuation and energy attenuation between them can be expressed as: in, ， and ; To prevent extremely small constants with a denominator of zero.

[0082] The physical eigenvector of the propagation attenuation branch can be represented as: in, 、 and Indicates the peak value of the three-channel disturbance; 、 and surface The disturbance energy of the three channels within the disturbance event window is shown. 、 、 Indicates the amplitude attenuation between channels; 、 、 This indicates the amount of energy attenuation between channels.

[0083] (3) Topological reflection branch The topology reflection branch is used to characterize reflection disturbances, secondary peaks, trailing waveforms, and phase disturbances caused by changes in equivalent impedance at branch nodes, load connection points, line ends, or power supply sides. This branch sets a reflection search window after the first arrival disturbance. in, This indicates the reflection search window following the initial disturbance. This represents the number of interval sampling points set to avoid first-arrival mutations; Indicates the endpoint of the reflection search window relative to The number of sampling points.

[0084] No. Each channel's reflection peak sampling points and Reflection peak It can be represented as: Reflection delay and reflection intensity ratio It can be represented as: in, surface Show the first Each channel's reflection peak sampling points; Indicates the first The reflection peak value of each channel; surface Show the first The reflection delay of each channel; Indicates the first The ratio of reflection intensity of each channel.

[0085] Line topology information Encoded as topological feature vectors: The physical eigenvectors of the topological reflection branch can be represented as: in, This represents the topology embedding vector obtained by encoding the line topology information.

[0086] （ 4) Slow-changing load branch The slow-change load branch is used to characterize load periodic fluctuations, power factor variations, equivalent impedance changes, and steady-state distortion trends. This branch uses energy metering parameters synchronously acquired by the energy metering and communication module. 。 Let the first The load state parameter vector within each electricity metering window is: in: : The effective value of the voltage in the k-th window; : The effective value of the current in the k-th window; Active power in the k-th window; : Reactive power in the k-th window; Apparent power of the k-th window; Power factor of the k-th window; ： Equivalent impedance of the load in the k-th window; : Load state parameter vector of the k-th window.

[0087] The load change rate can be expressed as: Encode the electricity metering parameters into a metering embedding vector: The physical eigenvector of the slowly varying load branch can be represented as: in, This indicates a load stability metric; This represents the metering embedding vector obtained by encoding the electricity metering parameters.

[0088] In one possible implementation, step S105 corrects the initial distance based on the residual sensitivity characteristics to obtain the corrected distance. This may include: determining the distance residual correction amount corresponding to line parameters, load status, calibration compensation, and temperature drift suppression based on the residual sensitivity characteristics and pre-acquired power metering parameters and temperature drift suppression parameters; and correcting the initial distance based on the distance residual correction amount corresponding to line parameters, load status, calibration compensation, and temperature drift suppression to obtain the corrected distance. Specific steps include: (1) Acquisition of electricity metering parameters First, obtain the energy metering parameters of the circuit under test, including but not limited to RMS voltage, RMS current, active power, reactive power, apparent power, power factor, load change rate, and equivalent impedance reference value. The set of energy metering parameters can be represented as follows: Electricity metering parameters are not directly used as the basis for determining whether a fault arc has occurred. Instead, they serve as auxiliary inputs for distance correction, interference state modeling, and reliability correction, reflecting the current load status and line operating status.

[0089] (2) Multi-parameter distance correction and temperature drift suppression based on residual sensitivity characteristics The set of temperature drift suppression parameters can be expressed as: in, Indicates the temperature of the sensor or measuring end. Indicates ambient temperature. This represents the estimated temperature rise of the line. Indicates reference calibration temperature ， Indicates the temperature offset. This represents the temperature drift suppression coefficient.

[0090] To explain the sources of various residual corrections, we first construct four types of correction input vectors: in, Represents the set of line parameters. Represents the set of electricity metering parameters. Indicates historical calibration compensation parameters. This represents the set of temperature drift suppression parameters. for The residual sensitivity of the dual-path Mamba output is used to characterize the correlation between the current initial distance error and line parameters, load status, calibration error, or temperature drift.

[0091] Various residual corrections can be expressed as: in, 、 、 and These represent the distance residual correction amounts corresponding to line parameters, load status, calibration compensation, and temperature drift suppression, respectively. 、 、 and This represents the mapping parameters for the corresponding correction branch; 、 、 and This represents the bias parameter for the corresponding correction branch. Used to limit the compensation magnitude of a single correction branch, to prevent a certain type of correction from being too large; 、 、 and surface The maximum permissible amplitude of various correction values ​​can be determined by historical calibration experiments or line parameter ranges.

[0092] The aforementioned residual correction amounts represent the distance that may need to be compensated for for each type of error source. To avoid simply superimposing the four types of corrections, this embodiment further dynamically determines the correction weights based on state reliability, disturbance signal-to-noise ratio, load stability, topology parameter integrity, historical calibration error, and temperature offset. The correction weight input vector can be expressed as: in, Indicates the reliability of the two-way Mamba state. This represents the perturbation signal-to-noise ratio. Indicates load stability. This indicates the completeness of topological parameters or the reliability of topological reflection. Indicates historical calibration error. This represents the absolute value of the temperature offset.

[0093] The scores for the four correction branches are generated using the corrected weight scoring function: If a linear scoring method is used, it can be expressed as: in, This represents the corrected weighted scoring function. and These represent the weighted scoring matrix and the bias vector, respectively. 、 、 and These represent the scores for four types of correction branches: line parameters, load status, calibration compensation, and temperature drift suppression.

[0094] The four correction weights are determined by the softmax function: Among them, the The class correction weights are: Therefore, the four correction weights satisfy: The correction weight indicates which type of residual compensation should be trusted more in the current state. When line parameter deviations are more sensitive to residuals... Increase; when load fluctuations are more pronounced. Increase; when the impact of historical calibration errors is more significant. Increase; when the temperature deviation is large. Increase.

[0095] The final correction distance can be expressed as: in, This indicates the corrected fault arc distance.

[0096] This process combines the residual sensitivity of the dual-path Mamba output with multi-parameter correction, enabling the correction module to determine the main source of the current ranging error and to provide targeted compensation for line parameter deviations, load state changes, historical calibration errors, and temperature drift.

[0097] In one possible implementation, step S105, based on the calibration distance and state reliability, determines the fault arc detection result, which may include: determining the distance reliability based on the state reliability and pre-acquired channel disturbance synchronization degree, disturbance signal-to-noise ratio, line parameter integrity, and load state stability indicators; determining the final error range based on the distance reliability and pre-acquired sampling period error, line calibration error, state uncertainty indicator, and minimum error lower limit; and generating the fault arc detection result based on the calibration distance, distance reliability, and final error range. Specific steps include: The final distance confidence level is expressed as: in, These are the weighting coefficients, and ;C_d Indicates the final distance confidence level. C_s Indicates the reliability of the two-way Mamba state. C_syncIndicates the reliability of multi-channel synchronization. C_snr This indicates the reliability of the perturbation signal-to-noise ratio. C_ line This indicates the completeness and reliability of the line parameters. C_load This indicates the reliability of the load status.

[0098] The final error range can be determined by considering distance reliability, sampling period error, line calibration error, state uncertainty, and minimum error lower limit. in, e_d Indicates the error range. e_min This represents the lower bound of the minimum error. e_samp This indicates the distance resolution error caused by the sampling period. e_cal This indicates the line calibration error. e_state This indicates the uncertainty of the two-path Mamba state. γ This represents the confidence level error amplification factor.

[0099] In one possible implementation, since the fault arc distance should be within the length of the measured line, the correction distance can be constrained based on the length range of the measured line before output. The final distance measurement result can include the correction distance, error range, distance reliability, distance interval, fault branch, fault occurrence time, and alarm signal, etc.

[0100] Finally, the power metering and communication module uploads the ranging results to the local terminal, remote monitoring platform, or power distribution management system, and can also output alarm signals based on the ranging results.

[0101] The embodiments of the present invention achieve the following beneficial effects: (i) Expanding from fault alarm to distance measurement improves fault location efficiency. This embodiment extracts distance-related features such as time difference of arrival, propagation attenuation, phase offset, disturbance energy, and rising edge slope, and further outputs the corrected distance, error range, distance interval, and distance reliability, providing a basis for maintenance personnel to quickly determine the location of the fault section and fault point.

[0102] (II) Improving the ranging targeting of multi-scale feature enhancement. This embodiment proposes a TimeMixer feature enhancement mechanism with propagation physical scale branch for fault arc ranging. It sets up a first-arrival abrupt branch, a propagation attenuation branch, a topological reflection branch, and a load slowly varying branch. The weights of each branch are determined based on channel propagation consistency, signal-to-noise ratio, load stability, and topological reflection reliability, so that the multi-scale enhancement process has a clear physical meaning of fault arc propagation.

[0103] (III) Improving ranging stability under complex lines. This embodiment uses a dual-path Mamba consisting of a ranging state path and an interference state path to selectively retain distance-related states such as first arrival, arrival difference, attenuation, and phase drift. It also models and suppresses load cycle fluctuations, random noise, switching transients, slow temperature drift, and non-ranging electromagnetic interference, avoiding mistaking interference states for distance states.

[0104] (iv) Improve the pertinence and reliability of distance correction. This embodiment uses the residual sensitivity characteristics of the dual-path Mamba output to determine the source of the initial distance error, and then combines line parameters, power metering parameters, historical calibration parameters and temperature drift suppression parameters to perform dynamic weighted residual compensation, thereby improving the reliability of distance measurement under different line lengths, branch topologies, load conditions and high temperature drift conditions.

[0105] (v) Improve the engineering usability of ranging results. In addition to outputting the corrected distance, this embodiment further provides the distance interval, error range, distance reliability and the branch where the fault is located, and imposes line length boundary constraints on the output results, making the ranging results more suitable for fault diagnosis and maintenance decisions.

[0106] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0107] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0108] Figure 4 A schematic diagram of the fault arc detection device provided in an embodiment of the present invention is shown. For ease of explanation, only the parts related to the embodiment of the present invention are shown, and are described in detail below: like Figure 4 As shown, the fault arc detection device includes: The multi-channel sensing module 41 is used to acquire various disturbance signals of the target power distribution line; The disturbance analysis module 42 is used to determine the disturbance analysis time window based on multiple types of disturbance signals, and to extract basic ranging-related features based on the multiple types of disturbance signals within the disturbance analysis time window; The feature enhancement module 43 is used to generate output features of each branch based on multiple types of disturbance signals through a preset neural network model corresponding to different fault arc propagation physical branches, and to fuse the output features of each branch with the basic ranging-related features to obtain an enhanced ranging feature vector; wherein, the preset neural network model corresponding to each branch has different time selection operators and time mixing blocks. The ranging module 44 is used to input the enhanced ranging feature vector into the preset ranging network and output the initial distance of the fault arc, residual sensitive features and state confidence. The correction module 45 is used to correct the initial distance based on the residual sensitivity characteristics to obtain the corrected distance, and to determine the fault arc detection result based on the corrected distance and the state confidence level.

[0109] Figure 5 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Figure 5 As shown, the electronic device 5 of this embodiment includes a processor 50 and a memory 51. The memory 51 stores a computer program 52. When the processor 50 executes the computer program 52, it implements the steps in the various method embodiments described above. Alternatively, when the processor 50 executes the computer program 52, it implements the functions of each module / unit in the various device embodiments described above.

[0110] For example, computer program 52 may be divided into one or more modules / units, which are stored in memory 51 and executed by processor 50 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 52 in electronic device 5.

[0111] Electronic device 5 may include, but is not limited to, processor 50 and memory 51. Those skilled in the art will understand that... Figure 5 This is merely an example of electronic device 5 and does not constitute a limitation on electronic device 5. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 5 may also include input / output devices, network access devices, buses, etc.

[0112] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0113] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0114] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method of detecting a fault arc, characterized by, include: Acquire various disturbance signals from the target power distribution line; Based on the various types of disturbance signals, a disturbance analysis time window is determined, and based on the various types of disturbance signals within the disturbance analysis time window, basic ranging correlation features are extracted. Based on the various types of disturbance signals, the output features of each branch are generated through the preset neural network model corresponding to different fault arc propagation physical branches. The output features of each branch and the basic ranging-related features are then fused to obtain an enhanced ranging feature vector. The preset neural network model corresponding to each branch has different time selection operators and time mixing blocks. The enhanced ranging feature vector is input into a preset ranging network, and the initial distance of the fault arc, residual sensitive features, and state reliability are output. The initial distance is corrected based on the residual sensitivity feature to obtain the corrected distance. Based on the corrected distance and the state confidence level, the fault arc detection result is determined.

2. The arc fault detection method of claim 1, wherein, The step of determining the disturbance analysis time window based on the multiple types of disturbance signals includes: Determine the combined disturbance intensity of the multiple types of disturbance signals; If at any time the overall disturbance intensity is greater than a preset intensity threshold, then monitor the duration for which the overall disturbance intensity is greater than the preset intensity threshold, and calculate the high-frequency energy ratio and channel consistency index of the multiple disturbance signals. If at least two of the duration, the high-frequency energy ratio, and the channel consistency index satisfy the corresponding fault arc judgment conditions, then that moment is determined as the disturbance arrival reference moment, and the disturbance analysis time window is determined based on the disturbance arrival reference moment.

3. The arc fault detection method of claim 1, wherein, The step involves generating output features for each branch of the fault arc propagation physical branch based on the multiple types of disturbance signals, using a preset neural network model, and fusing the output features of each branch with the basic ranging-related features to obtain an enhanced ranging feature vector, including: By embedding the multiple types of perturbation signals using a shared input, multi-channel features are obtained. The multi-channel features are input into the preset neural network model corresponding to each physical branch of the fault arc propagation to obtain the time-series features corresponding to each branch. Construct a preset physical feature vector corresponding to each physical branch of fault arc propagation, and fuse the preset physical feature vectors corresponding to each branch with the time series features to obtain the output features of each branch; The output features of each branch are fused to obtain the enhanced output features; The enhanced output features are fused with the basic ranging-related features to obtain the enhanced ranging feature vector.

4. The arc fault detection method of claim 1, wherein, The ranging network includes: a ranging state path, an interference state path, and an interference suppression gating; The step of inputting the enhanced ranging feature vector into a preset ranging network and outputting the initial distance, residual sensitive features, and state confidence of the fault arc includes: The enhanced ranging feature vector is input into the ranging state path to obtain the distance state feature; The enhanced ranging feature vector, along with the pre-acquired power metering parameters and temperature drift suppression parameters, are input into the interference state path to obtain the interference state features. The distance state feature and the interference state feature are input into the interference suppression gating so that the distance state feature is corrected by the interference state feature; The corrected distance state characteristics are input into three preset output heads to obtain the initial distance, residual sensitive characteristics, and state reliability of the fault arc output by the three output heads.

5. The fault arc detection method according to claim 1, characterized in that, The step of correcting the initial distance based on the residual sensitivity feature to obtain the corrected distance includes: Based on the residual sensitivity characteristics and the pre-acquired power metering parameters and temperature drift suppression parameters, determine the distance residual correction amount corresponding to the line parameters, load status, calibration compensation and temperature drift suppression; Based on the line parameters, the load status, the calibration compensation, and the distance residual correction amount corresponding to the temperature drift suppression, the initial distance is corrected to obtain the corrected distance.

6. The fault arc detection method according to claim 1, characterized in that, The determination of the fault arc detection result based on the correction distance and the state confidence level includes: Based on the state reliability, as well as the pre-acquired channel disturbance synchronization degree, disturbance signal-to-noise ratio, line parameter integrity and load state stability indicators, the distance reliability is determined; Based on the distance reliability, as well as the pre-acquired sampling period error, line calibration error, state uncertainty index, and minimum error lower limit, the final error range is determined; The fault arc detection result is generated based on the correction distance, the distance reliability, and the final error range.

7. The fault arc detection method according to any one of claims 1-6, characterized in that, The various types of disturbance signals include: current disturbance signals, voltage disturbance signals, and near-field electromagnetic disturbance signals.

8. The fault arc detection method according to any one of claims 1-6, characterized in that, The different physical branches of fault arc propagation include: The first-arrival mutation branch is used to reflect the mutation information generated when the fault arc first arrives; The propagation attenuation branch is used to reflect the attenuation information of the fault arc during its propagation along the line; Topological reflection branch is used to reflect the reflection changes caused by abrupt changes in the equivalent impedance of line nodes; The slow-change load branch is used to reflect changes in load status parameters.

9. A fault arc detection device, characterized in that, include: A multi-channel sensing module is used to acquire various disturbance signals of the target power distribution line; The disturbance analysis module is used to determine a disturbance analysis time window based on the multiple disturbance signals, and to extract basic ranging-related features based on the multiple disturbance signals within the disturbance analysis time window; The feature enhancement module is used to generate output features of each branch based on the multiple types of disturbance signals through a preset neural network model corresponding to different fault arc propagation physical branches, and to fuse the output features of each branch with the basic ranging-related features to obtain an enhanced ranging feature vector; wherein, the preset neural network model corresponding to each branch has different time selection operators and time mixing blocks. The ranging module is used to input the enhanced ranging feature vector into a preset ranging network and output the initial distance of the fault arc, residual sensitive features, and state reliability. The correction module is used to correct the initial distance according to the residual sensitive feature to obtain the corrected distance, and to determine the fault arc detection result based on the corrected distance and the state confidence level.

10. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 8.

Images