Power distribution overhead line fault traveling wave positioning anti-electromagnetic interference method and system and medium

Abstract

The invention relates to the technical field of electrical variable measurement, in particular to a power distribution overhead line fault traveling wave positioning anti-electromagnetic interference method and system and a medium. According to the method, a broadband equivalent circuit model of a power distribution overhead line is constructed based on a vector matching method and is used for simulating transmission characteristics of fault traveling waves and switching operation overvoltage; performing multi-scale decomposition on the acquired signals by adopting a wavelet transform algorithm, extracting disturbance signal parameters, and calculating traveling wave propagation characteristics according to different fault types and switch working conditions; constructing an electromagnetic disturbance propagation coupling model, and performing time-frequency domain analysis by taking the traveling wave propagation characteristics, the disturbance signal parameters and the line environment parameters as input to obtain coupling characteristics of electromagnetic disturbance; carrying out numerical simulation by utilizing electromagnetic field simulation software, analyzing the influence of line environment parameters on coupling characteristics, and optimizing model parameters; and finally, analyzing the coupling characteristics of electromagnetic disturbance, eliminating the influence of electromagnetic interference on the traveling wave signal through a reverse decoupling algorithm, and achieving the accurate fault positioning of the power distribution overhead line.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical variable measurement, and in particular to a method, system and medium for anti-electromagnetic interference traveling wave positioning of a distribution overhead line fault. Background Art

[0002] As a vital component of the power system, the distribution network is directly related to the reliability and quality of power supply to users. With the continuous expansion of power systems and the advancement of smart grid construction, rapid fault location technology for overhead distribution lines has become one of the key technologies to ensure the safe and stable operation of the power grid.

[0003] Traditional distribution line fault location methods primarily include impedance methods, traveling wave methods, and intelligent algorithms. The traveling wave method, however, has been widely used in distribution network fault location due to its unaffected system operating mode and high positioning accuracy. The basic principle of the traveling wave method is to utilize transient traveling wave signals generated during a fault, measure the time it takes for the traveling wave to reach the measurement point, and calculate the fault distance based on the traveling wave's propagation speed.

[0004] However, in practical applications, fault location using traveling waves on distribution overhead lines faces a serious challenge. Firstly, the distribution network contains numerous nonlinear loads, power electronic equipment, and switching devices, which generate various electromagnetic interferences during normal operation and during fault conditions. Secondly, the open structure of overhead lines makes them susceptible to external electromagnetic interference, such as lightning, wireless communications, and industrial equipment. These electromagnetic interferences overlap with the fault traveling wave signals in the frequency domain, making it difficult for traditional filtering methods to effectively distinguish and suppress them. This results in distortion of the fault traveling wave signals, severely impacting location accuracy. Summary of the Invention

[0005] In response to the deficiencies in the prior art, the present invention provides a method, system, and medium for anti-electromagnetic interference traveling wave positioning of distribution overhead line faults, which solves the problem that electromagnetic interference causes distortion of fault traveling wave signals, seriously affecting positioning accuracy.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for fault location of distribution overhead lines with traveling waves and anti-electromagnetic interference, comprising:

[0007] A broadband equivalent circuit model of the distribution overhead line is constructed based on the vector matching method to simulate the fault traveling wave and the transmission fault signal of the switching overvoltage.

[0008] The wavelet transform algorithm is used to perform multi-scale decomposition on the collected fault signal, distinguish the signal type through time-frequency domain characteristics, extract the disturbance signal parameters, and calculate the traveling wave propagation characteristics for different fault types and switching conditions;

[0009] Combining a long short-term memory neural network and a convolutional neural network to construct an electromagnetic disturbance propagation coupling model, using the traveling wave propagation characteristics, disturbance signal parameters, and line environment parameters as inputs and performing time-frequency domain analysis to obtain the coupling characteristics of the electromagnetic disturbance;

[0010] Use electromagnetic field simulation software to perform numerical simulations, analyze the impact of line environment parameters on coupling characteristics based on simulation data, and compare and verify with field experimental data to optimize the parameters of the electromagnetic disturbance propagation coupling model;

[0011] Based on the optimized model, the coupling characteristics of electromagnetic disturbance are analyzed, the influence of electromagnetic interference on traveling wave signals is eliminated, and fault location of distribution overhead lines is realized.

[0012] Furthermore, the steps for constructing a broadband equivalent circuit model are:

[0013] Obtain physical parameters of overhead distribution lines, including conductor radius, conductor spacing, height above ground, and line length;

[0014] Based on the multi-conductor transmission line theory, the frequency-dependent unit length parameter matrix of the line is calculated, including the resistance matrix, inductance matrix, conductance matrix, and capacitance matrix, and the frequency characteristics of the mutual inductance between phases and mutual inductance to ground are considered;

[0015] Calculate the impedance characteristics of the line at multiple frequency points in the frequency domain, covering the frequency range from 50Hz to 10MHz;

[0016] The vector matching algorithm is used to approximate the frequency domain impedance data with rational functions, and a transfer function model in the pole-residue form is established.

[0017] The transfer function model is converted into an equivalent circuit network, and the pole-residue circuit mapping is achieved through the combination of RC, RL, or RLC branches.

[0018] Furthermore, the steps of extracting disturbance signal parameters are as follows:

[0019] Select the mother wavelet function to perform multi-scale wavelet decomposition on the collected transmission fault signal, and decompose the signal into wavelet coefficients of different frequency bands;

[0020] Signal types are distinguished based on the energy distribution characteristics of wavelet coefficients. The high-frequency wavelet coefficients of fault traveling wave signals show a peak feature, while the medium- and low-frequency wavelet coefficients of switching operation overvoltage signals are dominant.

[0021] Extract the time domain parameters of the disturbance signal from the wavelet decomposition results, including signal amplitude, rise time, pulse width and decay time constant;

[0022] The frequency domain parameters of the disturbance signal are obtained by wavelet coefficient reconstruction, including main frequency component, spectrum bandwidth, frequency distribution and harmonic content;

[0023] Calculate the time-frequency domain joint characteristic parameters of wavelet transform, including instantaneous frequency, spectral center of gravity, signal singularity index and wavelet entropy.

[0024] Furthermore, the steps for calculating the traveling wave propagation characteristics are:

[0025] Determine the corresponding propagation mode based on the identified signal type, including the zero-sequence mode corresponding to a single-phase ground fault, and the positive-sequence and negative-sequence modes corresponding to a phase-to-phase short-circuit fault;

[0026] Calculate the reflection coefficient and refraction coefficient of the traveling wave based on the characteristic impedance and boundary conditions of the line;

[0027] Calculate the attenuation characteristics of traveling waves during propagation, including the attenuation caused by conductor loss and dielectric loss;

[0028] Based on the switching operation conditions, the overvoltage propagation characteristics generated by closing and opening operations are calculated, including the overvoltage amplitude and spectrum characteristics.

[0029] Furthermore, the electromagnetic disturbance propagation coupling model includes:

[0030] The input layer receives multi-dimensional input data, including traveling wave propagation characteristics, disturbance signal parameters, and line environment parameters;

[0031] The feature extraction layer uses a one-dimensional convolutional neural network combined with wavelet transform preprocessing to extract local features through convolution operations and apply activation functions to process nonlinearity, decomposing the input signal into multi-scale frequency bands;

[0032] The time series modeling layer uses a bidirectional long short-term memory network to control the flow of information through forget gates, input gates, and output gates. It also introduces an attention mechanism to dynamically weight the coupling signal features at different time steps and output the propagation attenuation coefficient and coupling gain of the electromagnetic disturbance.

[0033] The output layer generates a coupling characteristic matrix containing multiple predicted location points and corresponding coupling characteristic indicators through the fully connected layer.

[0034] Furthermore, the steps of numerical simulation using electromagnetic field simulation software are as follows:

[0035] High-frequency sensors are used to collect fault traveling wave signals from overhead distribution lines. A spectrum analyzer is used to measure the switching overvoltage waveform and electromagnetic disturbance spectrum. Electromagnetic field simulation software is also used to generate synthetic data under different operating conditions.

[0036] The collected sample data is normalized and the time-frequency feature data required for training is generated through time-frequency transformation.

[0037] Furthermore, the steps for optimizing the parameters of the electromagnetic disturbance propagation coupling model are as follows:

[0038] A transfer learning strategy is used for model training. The model is first pre-trained on simulation data to learn the general electromagnetic interference coupling mechanism, and then the model weights are fine-tuned on real data to adapt to the actual line characteristics.

[0039] Use composite loss function for model optimization, combining regression loss and classification loss to achieve multi-objective optimization;

[0040] Early stopping and cross-validation methods are used during training to evaluate model performance, prevent overfitting, and improve generalization capabilities.

[0041] An online adaptive update mechanism is established. When the model prediction deviation exceeds the set threshold, the retraining process is triggered and the model parameters are updated using the newly collected data.

[0042] Furthermore, the steps for locating the fault of the distribution overhead line are:

[0043] Use the trained electromagnetic disturbance propagation coupling model to analyze the real-time collected traveling wave signal and obtain the coupling characteristic matrix;

[0044] The original traveling wave signal is subjected to interference suppression processing based on the coupling characteristic matrix, and the pure fault traveling wave component is separated from the mixed signal through the reverse decoupling algorithm;

[0045] The purified fault traveling wave signal is used to calculate the traveling wave propagation time and reflection characteristics, and the fault location is determined in combination with the line propagation characteristics.

[0046] The present invention also provides a system for locating faults of distribution overhead lines with traveling waves and resisting electromagnetic interference, which is used to implement the above-mentioned method for locating faults of distribution overhead lines with traveling waves and resisting electromagnetic interference, comprising:

[0047] The equivalent model building module is used to construct a broadband equivalent circuit model of the distribution overhead line based on the vector matching method, and is used to simulate the transmission fault signals of fault traveling waves and switching overvoltage;

[0048] The fault signal analysis module uses a wavelet transform algorithm to perform multi-scale decomposition on the collected transmission fault signal, distinguish the signal type through time-frequency domain characteristics, extract the disturbance signal parameters, and calculate the traveling wave propagation characteristics for different fault types and switching conditions;

[0049] A coupling model building module is used to combine a long short-term memory neural network and a convolutional neural network to build an electromagnetic disturbance propagation coupling model, taking the traveling wave propagation characteristics, disturbance signal parameters and line environment parameters as input and performing time-frequency domain analysis to obtain the coupling characteristics of the electromagnetic disturbance;

[0050] The coupling model optimization module is used to perform numerical simulation using electromagnetic field simulation software, analyze the impact of line environment parameters on coupling characteristics based on simulation data, and compare and verify with field experimental data to optimize the parameters of the electromagnetic disturbance propagation coupling model;

[0051] The fault location module is used to analyze the coupling characteristics of electromagnetic disturbances based on the optimized model, eliminate the impact of electromagnetic interference on traveling wave signals, and realize fault location of distribution overhead lines.

[0052] The present invention also provides a readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the above-mentioned method for fault location and electromagnetic interference resistance of distribution overhead lines using traveling waves is implemented.

[0053] The present invention has the following beneficial effects:

[0054] 1. This paper combines vector matching with deep learning technology to construct a broadband equivalent circuit model covering frequencies from 50 Hz to 10 MHz. This model accurately describes the frequency-dependent parameter characteristics of overhead distribution lines, including the frequency response of impedance, admittance, and mutual inductance. Combining the multi-scale decomposition of wavelet transforms with the feature extraction capabilities of neural networks, it enables precise differentiation and classification of fault traveling waves and switching overvoltage signals.

[0055] 2. This paper proposes an electromagnetic disturbance propagation coupling model based on a hybrid LSTM and CNN architecture. This model uses a convolutional neural network to extract local features, a long-short-term memory network to model temporal dependencies, and an attention mechanism to achieve dynamic weight allocation. This model can accurately predict the propagation attenuation coefficient and coupling gain of electromagnetic disturbances in distribution lines. By combining a transfer learning strategy with pre-training on simulation data and fine-tuning on real data, the model possesses strong generalization capabilities and adaptive optimization characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 Flowchart of the traveling wave fault location and electromagnetic interference resistance method for distribution overhead lines provided by the present invention;

[0057] Figure 2 A flowchart of the steps for constructing a broadband equivalent circuit provided by the present invention;

[0058] Figure 3 A flowchart of the steps for extracting disturbance signal parameters provided by the present invention;

[0059] Figure 4 A flow chart of the steps for calculating the traveling wave propagation characteristics provided by the present invention;

[0060] Figure 5 This is a structural diagram of the traveling wave fault location and anti-electromagnetic interference system for distribution overhead lines provided by the present invention. DETAILED DESCRIPTION

[0061] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0062] The embodiment of the present invention provides a method for fault location of distribution overhead lines with traveling waves and anti-electromagnetic interference, such as Figure 1 Shown, including:

[0063] S100: Constructs a broadband equivalent circuit model of overhead distribution lines based on the vector matching method to simulate fault traveling waves and switching overvoltage transmission fault signals;

[0064] Furthermore, the steps of constructing a broadband equivalent circuit model are as follows: Figure 2 Shown, including:

[0065] Obtain physical parameters of overhead distribution lines, including conductor radius, conductor spacing, height above ground, and line length;

[0066] Based on the multi-conductor transmission line theory, the frequency-dependent unit length parameter matrix of the line is calculated, including the resistance matrix, inductance matrix, conductance matrix, and capacitance matrix, and the frequency characteristics of the mutual inductance between phases and mutual inductance to ground are considered;

[0067] Calculate the impedance characteristics of the line at multiple frequency points in the frequency domain, covering the frequency range from 50Hz to 10MHz;

[0068] The vector matching algorithm is used to approximate the frequency domain impedance data with rational functions, and a transfer function model in the pole-residue form is established.

[0069] The transfer function model is converted into an equivalent circuit network, and the pole-residue circuit mapping is achieved through the combination of RC, RL, or RLC branches.

[0070] Specifically, physical structural parameters of overhead distribution lines are collected, including: conductor radius (r), used to calculate capacitance and resistance; conductor spacing (d), which determines coupling strength and mutual inductance; height above ground (h), which affects ground capacitance and zero-sequence component propagation; and line length (l), used for subsequent propagation path modeling and simulation boundary setting. These parameters can be obtained through field measurements, design drawings, or GIS data.

[0071] Based on the multi-conductor transmission line theory, a unit length transmission line model is established to calculate its frequency-related electrical parameters, including resistance matrix, inductance matrix, conductance matrix and capacitance matrix. The range is 50Hz to 10MHz. Resistor Matrix The skin effect formula modified by Bessel function is used for calculation; the inductance matrix and capacitor matrix Calculated by Carson formula and mirror method, considering the frequency characteristics of phase mutual inductance and mutual inductance to ground; conductivity matrix Calculated based on the dielectric constant and loss tangent of the insulating medium.

[0072] At the set frequency points (such as 50Hz, 100Hz, 1kHz, 10kHz, 100kHz, 1MHz, 10MHz, etc.), the frequency domain impedance matrix of the line is constructed according to the following formula:

[0073] ;

[0074] Where j represents the imaginary unit, and 2πf represents the angular frequency in frequency domain analysis. This impedance characteristic is used to characterize the line's response to transient signals such as traveling waves and switching overvoltage at different frequencies.

[0075] The frequency domain impedance data are fitted with rational functions using a vector matching algorithm, and a transfer function model of the following form is constructed:

[0076] ;

[0077] in, , represents the complex frequency domain variable, Indicates to keep some, Indicates the extreme point, 、 represents the DC term and the slope term, Indicates the order of matching, usually 5~20.

[0078] Based on the transfer function model in the form of pole-residue, a physical equivalent circuit network is established. Each first-order pole term in the mapping relationship It can be mapped into an RC or RL branch. The complex conjugate pole pairs can be combined into an RLC series / parallel network to simulate oscillation behavior. Multiple branches are connected in parallel to form a π-type or T-type network structure for overall modeling. The output circuit can be used for transient response analysis in time domain simulators (such as EMTP / PSCAD).

[0079] The broadband equivalent circuit model constructed in the above way retains the frequency dependence of line parameters, improves modeling accuracy, can distinguish between fault traveling waves and high-frequency signals such as switching overvoltage, and can be used for electromagnetic simulation and neural network model fusion input to improve prediction accuracy.

[0080] S200: Uses a wavelet transform algorithm to perform multi-scale decomposition on the collected transmission fault signal, distinguishes signal types based on time-frequency domain characteristics, extracts disturbance signal parameters, and calculates traveling wave propagation characteristics for different fault types and switching conditions.

[0081] Furthermore, the steps of extracting disturbance signal parameters are as follows: Figure 3 Shown, including:

[0082] Select the mother wavelet function to perform multi-scale wavelet decomposition on the collected transmission fault signal, and decompose the signal into wavelet coefficients of different frequency bands;

[0083] Signal types are distinguished based on the energy distribution characteristics of wavelet coefficients. The high-frequency wavelet coefficients of fault traveling wave signals show a peak feature, while the medium- and low-frequency wavelet coefficients of switching operation overvoltage signals are dominant.

[0084] Extract the time domain parameters of the disturbance signal from the wavelet decomposition results, including signal amplitude, rise time, pulse width and decay time constant;

[0085] The frequency domain parameters of the disturbance signal are obtained by wavelet coefficient reconstruction, including main frequency component, spectrum bandwidth, frequency distribution and harmonic content;

[0086] Calculate the time-frequency domain joint characteristic parameters of wavelet transform, including instantaneous frequency, spectral center of gravity, signal singularity index and wavelet entropy.

[0087] Specifically, according to the transient characteristics and time-frequency distribution characteristics of the processed signal, a suitable mother wavelet function is selected, preferably a wavelet basis function with good localization properties such as Daubechies (db4), Symlet or Coiflet. Perform multi-scale wavelet decomposition to obtain the wavelet coefficient sequence:

[0088] ;

[0089] in, represents the decomposition scale, Indicates the Layer k-th detail coefficient (high frequency component), Indicates the The kth approximation coefficient of the layer (low-frequency component), 、 Represents the wavelet function and scaling function. These two functions are commonly used in signal analysis, corresponding to the orthogonal basis functions for detail information and approximation information, respectively. Typically, the decomposition layer is 4 to 6, covering a frequency range of 50 Hz to 10 MHz.

[0090] For the obtained wavelet coefficients , calculate the energy distribution of each layer:

[0091] ;

[0092] The signal type is initially determined based on the energy proportion of each layer. If the high-frequency layer (such as layers 1 to 2) has a significant energy peak and a spike structure, it is determined to be a fault traveling wave. If the energy of the medium and low-frequency layers (such as layers 3 to 5) is dominant, it is determined to be a switching operation overvoltage signal.

[0093] Based on the reconstructed signals at each scale from the wavelet decomposition results, time-domain characteristic parameters are extracted, including but not limited to amplitude (maximum signal amplitude); rise time (time required for the signal to rise from 10% to 90% of its peak value); pulse width (duration over which the signal exceeds 50% of its peak value); and decay time constant (time required for the signal energy to decay to 1 / e of its initial value). These parameters are used as input to the subsequent coupling model and for fault propagation velocity estimation.

[0094] Through spectral reconstruction of wavelet coefficients (i.e., short-time Fourier transform), frequency domain characteristic parameters are extracted, including the dominant frequency component (the frequency with the highest energy concentration in the signal); the spectral bandwidth (the frequency range containing 90% of the energy); the frequency distribution (single-peak, double-peak, multi-peak, etc.); and the harmonic content (the proportion of non-fundamental components in the spectrum). These characteristics can be used to assess the transmission risk and coupling strength of interfering signals.

[0095] Furthermore, combining the time-frequency characteristics of wavelets, a series of discriminative joint characteristic parameters are calculated, including instantaneous frequency, which reflects the frequency variation of the signal at each time point; spectral centroid, which is the center of gravity of the signal's spectral distribution; singularity index, which measures the degree of signal mutation and is used to identify nonlinear shocks; and wavelet entropy, which reflects the complexity of the wavelet energy distribution and measures signal uncertainty. These parameters can be used as inputs to neural network models or independently for interference type determination.

[0096] Through the above-mentioned wavelet transform and multidimensional feature extraction process, different types of electromagnetic interference can be accurately identified from the mixed signal, and high-quality input data can be provided for subsequent propagation modeling, neural network training and fault location, thereby improving the adaptability of the entire system to high-frequency interference and nonlinear signals.

[0097] Furthermore, the steps for calculating the traveling wave propagation characteristics are as follows: Figure 4 Shown, including:

[0098] Determine the corresponding propagation mode based on the identified signal type, including the zero-sequence mode corresponding to a single-phase ground fault, and the positive-sequence and negative-sequence modes corresponding to a phase-to-phase short-circuit fault;

[0099] Calculate the reflection coefficient and refraction coefficient of the traveling wave based on the characteristic impedance and boundary conditions of the line;

[0100] Calculate the attenuation characteristics of traveling waves during propagation, including the attenuation caused by conductor loss and dielectric loss;

[0101] Based on the switching operation conditions, the overvoltage propagation characteristics generated by closing and opening operations are calculated, including the overvoltage amplitude and spectrum characteristics.

[0102] Specifically, the appropriate propagation mode is selected based on the signal type (fault traveling wave or switching overvoltage) identified through wavelet analysis and its three-phase symmetry. Single-phase grounding faults primarily propagate in the zero-sequence mode, with the signal reflecting and propagating along the ground channel, resulting in significant low-frequency losses. Phase-to-phase short-circuit faults involve symmetrical components, requiring consideration of both positive- and negative-sequence modes. For mixed faults, the Clarke transform or sequence component analysis method can be used to extract the composite propagation mode. The propagation velocity and impedance in each mode vary and must be calculated separately.

[0103] When there are discontinuities in the line (such as fault points, terminals, and transformer joints), it is necessary to calculate the reflection and transmission (refraction) characteristics of the signal at the interface. The reflection coefficient is defined using characteristic impedance and load impedance. ,in Represents the terminal load or interface impedance, Indicates the characteristic impedance of the line. In addition, the refractive index These two coefficients are used to determine whether the traveling wave is strongly reflected, weakly reflected, or propagates through the fault point or grounding point, which has a significant impact on subsequent positioning.

[0104] Due to the conductor loss and dielectric loss in the line, the amplitude of the traveling wave will be attenuated by different mechanisms during the propagation process. The first is the conductor loss attenuation, which is caused by resistance and mainly affects the high-frequency component. The calculation formula is ,in Represents the resistance per unit length. The second is dielectric loss attenuation, which is caused by media such as air and insulator strings, and is especially enhanced in rainy and foggy environments. The calculation formula is , where the angular velocity , represents the capacitance per unit length, represents the dielectric loss tangent, represents the characteristic admittance.

[0105] When the system closes or opens, it is usually accompanied by a spike overvoltage, and its propagation mode and spectrum characteristics also need to be calculated. During the closing operation, the capacitive branch is not fully charged, which is prone to closing inrush current. The closing overvoltage spectrum range is relatively wide, with the main frequency often ranging from 1kHz to 100kHz, and the peak amplitude is often 1.5 to 2.5 times the power frequency voltage. Typical spectrum templates can be extracted through EMTP / PSCAD simulation. The opening operation may cause arc reignition or capacitor residual voltage release, resulting in high-frequency resonant overvoltage. The main frequency may reach the MHz level. The opening overvoltage signal usually has an oscillating attenuation characteristic and can be expressed as:

[0106] ;

[0107] in, represents the initial overvoltage amplitude, represents the exponential decay term, represents the fundamental frequency of oscillation, The initial phase of the closing or opening state relative to the oscillation period is represented by . These parameters are used to establish the disturbance propagation model and serve as prior conditions or input features in the neural network.

[0108] Using this method, the propagation behavior of different types of fault traveling wave signals or switching overvoltage signals in overhead distribution lines can be accurately characterized, including mode classification (zero-sequence, positive-sequence, and negative-sequence), boundary reflection and transmission, distance attenuation and spectral distortion, and switching disturbance propagation paths. This provides high-quality propagation parameters for subsequent neural network training, interference modeling, and fault location.

[0109] S300: Constructing an electromagnetic disturbance propagation coupling model by combining a long short-term memory neural network and a convolutional neural network, taking the traveling wave propagation characteristics, disturbance signal parameters, and line environment parameters as inputs and performing time-frequency domain analysis to obtain coupling characteristics of the electromagnetic disturbance;

[0110] Furthermore, the electromagnetic disturbance propagation coupling model includes:

[0111] The input layer receives multi-dimensional input data, including traveling wave propagation characteristics, disturbance signal parameters, and line environment parameters;

[0112] The feature extraction layer uses a one-dimensional convolutional neural network combined with wavelet transform preprocessing to extract local features through convolution operations and apply activation functions to process nonlinearity, decomposing the input signal into multi-scale frequency bands;

[0113] The time series modeling layer uses a bidirectional long short-term memory network to control the flow of information through forget gates, input gates, and output gates. It also introduces an attention mechanism to dynamically weight the coupling signal features at different time steps and output the propagation attenuation coefficient and coupling gain of the electromagnetic disturbance.

[0114] The output layer generates a coupling characteristic matrix containing multiple predicted location points and corresponding coupling characteristic indicators through the fully connected layer.

[0115] Specifically, the input layer receives multi-dimensional input data extracted from the previous analysis, and the input dimension is set to ,in Indicates the number of input parameter types, which can be around 10 to 20. These parameters include, but are not limited to, traveling wave propagation characteristics: reflection coefficient, wave impedance, and propagation velocity; disturbance signal parameters: main frequency, bandwidth, amplitude, rise time, and wavelet entropy; and line environment parameters: conductor spacing, height above the ground, soil resistivity, and ambient noise level. Indicates the number of time series sampling points. For example, if sampling is performed at 1kHz, 1 second would be 1000 points. All input data is normalized before being fed into the network to improve training stability.

[0116] The feature extraction layer uses a one-dimensional convolutional neural network to extract local features from the input time series data. The convolution kernel parameters include kernel size (3×3 or 5×5), number of convolution layers (3 to 5), and number of channels per layer (16 to 64). The nonlinear activation layer of the convolution network uses the Reinforced Luminance (ReLU) function. A wavelet transform module is embedded before convolution to perform multi-scale frequency band decomposition on the input signal, extracting energy features from different frequency bands within the 50Hz to 10MHz range to improve sensitivity and resolution to high-frequency interference.

[0117] The temporal modeling layer uses a bidirectional long short-term memory network to dynamically model the temporal features of the convolutional output, capturing the coupled propagation characteristics of interference signals in the temporal dimension. The number of hidden units in the long short-term memory network ranges from 128 to 256, and a bidirectional structure is used for directionality to enhance context capture. A weighted attention mechanism is introduced to highlight the importance of key time points. The weights are calculated as follows:

[0118] ;

[0119] This formula can be interpreted as , first pass right Weight matrix and the bias vector The linear transformation of After the function is activated, the query vector is used Project it into an attention score , this step can be expressed as , and finally all Convert to normalized weights through softmax .

[0120] The output layer uses a fully connected network structure to map the time series features output by LSTM into a structured coupling characteristic matrix. The output dimension is ,in Indicates the predicted route location points, such as starting point, middle point, and end point. The output indicates the coupling characteristic indicators (such as amplitude, phase, main frequency, bandwidth, coupling gain, etc.) corresponding to each location point, usually 5 to 10. The output content predicts the propagation intensity of electromagnetic interference at each location point, providing reference values ​​for subsequent interference compensation and decoupling.

[0121] The electromagnetic interference propagation coupling model combines CNN and LSTM to simultaneously capture the spatial local characteristics and temporal dependencies of the signal. Wavelet transform embedding improves the ability to resolve high-frequency interference, and the attention mechanism increases sensitivity to key coupling positions and moments. The output is a structured characteristic matrix, which facilitates subsequent interference compensation, traveling wave decoupling, and fault location processing.

[0122] S400: Use electromagnetic field simulation software to perform numerical simulations, analyze the impact of line environment parameters on coupling characteristics based on simulation data, compare and verify with field experimental data, and optimize the parameters of the electromagnetic disturbance propagation coupling model;

[0123] Furthermore, the steps of numerical simulation using electromagnetic field simulation software are as follows:

[0124] High-frequency sensors are used to collect fault traveling wave signals from overhead distribution lines. A spectrum analyzer is used to measure the switching overvoltage waveform and electromagnetic disturbance spectrum. Electromagnetic field simulation software is also used to generate synthetic data under different operating conditions.

[0125] The collected sample data is normalized and the time-frequency feature data required for training is generated through time-frequency transformation.

[0126] Specifically, the model sample collection process begins with the acquisition of fault traveling wave signals from overhead distribution lines using high-frequency sensors at a sampling rate of at least 1 MHz. A spectrum analyzer is used to measure switching overvoltage waveforms and electromagnetic disturbance spectra. Simultaneously, synthetic data is generated using electromagnetic field simulation software such as Ansys HFSS to simulate various operating conditions, including line lengths ranging from 10 to 50 km, various fault types, and varying environmental parameters. The total number of collected samples is no less than 10,000, with approximately 30% real data and 70% synthetic data to ensure dataset diversity and coverage. Preprocessing includes normalization operations such as Min-Max Scaling to scale the data to the [0, 1] range and short-time Fourier transforms to generate time-frequency spectrograms. This generates a training dataset, 80% of which is used for training and 20% for validation.

[0127] By introducing the fusion training of high-fidelity electromagnetic simulation data and real measurement data, not only the model's adaptability to environmental changes is enhanced, but also the problems of "insufficient labeled samples" and "weak generalization ability" in traditional models are effectively solved.

[0128] Furthermore, the steps for optimizing the parameters of the electromagnetic disturbance propagation coupling model are as follows:

[0129] A transfer learning strategy is used for model training. The model is first pre-trained on simulation data to learn the general electromagnetic interference coupling mechanism, and then the model weights are fine-tuned on real data to adapt to the actual line characteristics.

[0130] Use composite loss function for model optimization, combining regression loss and classification loss to achieve multi-objective optimization;

[0131] Early stopping and cross-validation methods are used during training to evaluate model performance, prevent overfitting, and improve generalization capabilities.

[0132] An online adaptive update mechanism is established. When the model prediction deviation exceeds the set threshold, the retraining process is triggered and the model parameters are updated using the newly collected data.

[0133] Specifically, the Adam optimizer is used for model training, the initial learning rate is 0.001 and an exponential decay strategy is adopted, and the loss function is the mean square error Combined with cross entropy loss , total loss ,in The weight coefficient is about 0.1; the number of training iterations is 500 to 1000 rounds, the batch size is 32, and transfer learning is innovatively introduced. The model is first pre-trained on synthetic data to learn the general coupling mechanism, and iterates for about 300 rounds until The training process monitors accuracy and convergence, employing an early stopping mechanism to stop if the validation loss shows no improvement after 10 consecutive training rounds. After completion, the training process uses real-time signals and outputs coupling characteristic predictions. Parameters are optimized by comparing the predictions with experimental results. If the prediction deviation exceeds a threshold, such as 10%, online retraining is triggered, and the model is updated using newly collected data to achieve adaptive application. Furthermore, K-fold cross-validation (K=5) is used to assess model generalization.

[0134] By introducing a transfer learning strategy and a composite loss function, the model's ability to model the coupling characteristics of multi-source electromagnetic interference is effectively improved, enabling it to accurately predict coupling strength and identify interference types. Combining an early stopping mechanism with cross-validation helps prevent overfitting and enhances the model's generalization capabilities across diverse line environments.

[0135] S500: Analyzes the coupling characteristics of electromagnetic disturbances based on the optimized model, eliminates the impact of electromagnetic interference on traveling wave signals, and implements fault location on overhead distribution lines.

[0136] Furthermore, the steps for locating the fault of the distribution overhead line are:

[0137] Use the trained electromagnetic disturbance propagation coupling model to analyze the real-time collected traveling wave signal and obtain the coupling characteristic matrix;

[0138] The original traveling wave signal is subjected to interference suppression processing based on the coupling characteristic matrix, and the pure fault traveling wave component is separated from the mixed signal through the reverse decoupling algorithm;

[0139] The purified fault traveling wave signal is used to calculate the traveling wave propagation time and reflection characteristics, and the fault location is determined in combination with the line propagation characteristics.

[0140] Specifically, a high-frequency traveling wave detection device is installed at a substation or in the middle of a line to collect three-phase high-frequency voltage or current signals in real time. The signal sampling frequency is generally set between 5 and 20 MHz to cover the leading edge of the traveling wave. The collected signal sequence is input into a trained electromagnetic disturbance propagation coupling model, which outputs a coupling characteristic matrix corresponding to the time series.

[0141] Based on the above coupling characteristic matrix, the reverse decoupling matrix is ​​constructed , decoupling the original signal, the processing process can be expressed as: ,in represents the collected traveling wave signal, represents the pure fault traveling wave component, Represents the inverse matrix of the reverse decoupling matrix. This operation is equivalent to suppressing electromagnetic interference on the signal, removing irrelevant frequency bands and spurious disturbance components.

[0142] Wavelet transforms, high-frequency mutation detection, maximum slope point method, and energy threshold method can be used to extract the arrival time of the first wave head and analyze the characteristics of the reflected wave. Based on the boundary conditions at the end of the line (such as open circuit or short circuit), the presence of a reflected wave is detected. The time difference between the first wave head and its reflected wave is extracted. Based on the line propagation characteristics, the fault location is calculated using the formula: ,in Indicates the distance between the fault point and the measuring point, is the speed of traveling wave propagation, Indicates the time difference between the wave head and the reflected wave.

[0143] Through these steps, the system can extract the true fault component from contaminated traveling wave signals in strong electromagnetic interference environments and accurately restore the propagation path and reflection characteristics, significantly improving the accuracy of fault location under electromagnetic interference. Furthermore, by using a neural network to predict interference characteristics and participate in decoupling, it achieves dynamic compensation for multi-source nonlinear interference, resolving the issues that traditional methods face, such as susceptibility to noise interference, inability to locate faults, and even misjudgment.

[0144] The embodiment of the present invention also provides a traveling wave fault location and electromagnetic interference resistance system for distribution overhead lines, such as Figure 5 Shown, including:

[0145] The equivalent model building module is used to construct a broadband equivalent circuit model of the distribution overhead line based on the vector matching method, and is used to simulate the transmission fault signals of fault traveling waves and switching overvoltage;

[0146] The fault signal analysis module uses a wavelet transform algorithm to perform multi-scale decomposition on the collected transmission fault signal, distinguish the signal type through time-frequency domain characteristics, extract the disturbance signal parameters, and calculate the traveling wave propagation characteristics for different fault types and switching conditions;

[0147] A coupling model building module is used to combine a long short-term memory neural network and a convolutional neural network to build an electromagnetic disturbance propagation coupling model, taking the traveling wave propagation characteristics, disturbance signal parameters and line environment parameters as input and performing time-frequency domain analysis to obtain the coupling characteristics of the electromagnetic disturbance;

[0148] The coupling model optimization module is used to perform numerical simulation using electromagnetic field simulation software, analyze the impact of line environment parameters on coupling characteristics based on simulation data, and compare and verify with field experimental data to optimize the parameters of the electromagnetic disturbance propagation coupling model;

[0149] The fault location module is used to analyze the coupling characteristics of electromagnetic disturbances based on the optimized model, eliminate the impact of electromagnetic interference on traveling wave signals, and realize fault location of distribution overhead lines.

[0150] An embodiment of the present invention further provides a readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the above-mentioned method for fault location and electromagnetic interference resistance of distribution overhead lines using traveling waves is implemented.

[0151] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for locating faults in distribution overhead lines using traveling waves and resisting electromagnetic interference, characterized in that: include: A broadband equivalent circuit model of the distribution overhead line is constructed based on the vector matching method to simulate the fault traveling wave and the transmission fault signal of the switching overvoltage. The wavelet transform algorithm is used to perform multi-scale decomposition on the collected fault signal, distinguish the signal type through time-frequency domain characteristics, extract the disturbance signal parameters, and calculate the traveling wave propagation characteristics for different fault types and switching conditions; Combining a long short-term memory neural network and a convolutional neural network to construct an electromagnetic disturbance propagation coupling model, using the traveling wave propagation characteristics, disturbance signal parameters, and line environment parameters as inputs and performing time-frequency domain analysis to obtain the coupling characteristics of the electromagnetic disturbance; Use electromagnetic field simulation software to perform numerical simulations, analyze the impact of line environment parameters on coupling characteristics based on simulation data, and compare and verify with field experimental data to optimize the parameters of the electromagnetic disturbance propagation coupling model; Based on the optimized model, the coupling characteristics of electromagnetic disturbance are analyzed, the influence of electromagnetic interference on traveling wave signals is eliminated, and fault location of distribution overhead lines is realized.

2. The method for fault location and electromagnetic interference resistance of distribution overhead lines according to claim 1, characterized in that: The steps to construct a broadband equivalent circuit model are: Obtain physical parameters of overhead distribution lines, including conductor radius, conductor spacing, height above ground, and line length; Based on the multi-conductor transmission line theory, the frequency-dependent unit length parameter matrix of the line is calculated, including the resistance matrix, inductance matrix, conductance matrix, and capacitance matrix, and the frequency characteristics of the mutual inductance between phases and mutual inductance to ground are considered; Calculate the impedance characteristics of the line at multiple frequency points in the frequency domain, covering the frequency range from 50Hz to 10MHz; The vector matching algorithm is used to approximate the frequency domain impedance data with rational functions, and a transfer function model in the pole-residue form is established. The transfer function model is converted into an equivalent circuit network, and the pole-residue circuit mapping is achieved through the combination of RC, RL, or RLC branches.

3. The method for fault location and electromagnetic interference resistance of distribution overhead lines according to claim 1, characterized in that: The steps to extract disturbance signal parameters are as follows: Select the mother wavelet function to perform multi-scale wavelet decomposition on the collected transmission fault signal, and decompose the signal into wavelet coefficients of different frequency bands; Signal types are distinguished based on the energy distribution characteristics of wavelet coefficients. The high-frequency wavelet coefficients of fault traveling wave signals show a peak feature, while the medium- and low-frequency wavelet coefficients of switching operation overvoltage signals are dominant. Extract the time domain parameters of the disturbance signal from the wavelet decomposition results, including signal amplitude, rise time, pulse width and decay time constant; The frequency domain parameters of the disturbance signal are obtained by wavelet coefficient reconstruction, including main frequency component, spectrum bandwidth, frequency distribution and harmonic content; Calculate the time-frequency domain joint characteristic parameters of wavelet transform, including instantaneous frequency, spectral center of gravity, signal singularity index and wavelet entropy.

4. The method for fault location and electromagnetic interference resistance of distribution overhead lines according to claim 1, characterized in that: The steps to calculate the traveling wave propagation characteristics are: Determine the corresponding propagation mode based on the identified signal type, including the zero-sequence mode corresponding to a single-phase ground fault, and the positive-sequence and negative-sequence modes corresponding to a phase-to-phase short-circuit fault; Calculate the reflection coefficient and refraction coefficient of the traveling wave based on the characteristic impedance and boundary conditions of the line; Calculate the attenuation characteristics of traveling waves during propagation, including the attenuation caused by conductor loss and dielectric loss; Based on the switching operation conditions, the overvoltage propagation characteristics generated by closing and opening operations are calculated, including the overvoltage amplitude and spectrum characteristics.

5. The method for locating faults of distribution overhead lines using traveling waves and resisting electromagnetic interference according to claim 1, characterized in that: The electromagnetic disturbance propagation coupling model includes: The input layer receives multi-dimensional input data, including traveling wave propagation characteristics, disturbance signal parameters, and line environment parameters; The feature extraction layer uses a one-dimensional convolutional neural network combined with wavelet transform preprocessing to extract local features through convolution operations and apply activation functions to process nonlinearity, decomposing the input signal into multi-scale frequency bands; The time series modeling layer uses a bidirectional long short-term memory network to control the flow of information through forget gates, input gates, and output gates. It also introduces an attention mechanism to dynamically weight the coupling signal features at different time steps and output the propagation attenuation coefficient and coupling gain of the electromagnetic disturbance. The output layer generates a coupling characteristic matrix containing multiple predicted location points and corresponding coupling characteristic indicators through the fully connected layer.

6. The method for fault location and electromagnetic interference resistance of distribution overhead lines according to claim 1, characterized in that: The steps for numerical simulation using electromagnetic field simulation software are as follows: High-frequency sensors are used to collect fault traveling wave signals from overhead distribution lines. A spectrum analyzer is used to measure the switching overvoltage waveform and electromagnetic disturbance spectrum. Electromagnetic field simulation software is also used to generate synthetic data under different operating conditions. The collected sample data is normalized and the time-frequency feature data required for training is generated through time-frequency transformation.

7. The method for locating faults of distribution overhead lines using traveling waves and resisting electromagnetic interference according to claim 1, characterized in that: The steps to optimize the parameters of the electromagnetic disturbance propagation coupling model are: A transfer learning strategy is used for model training. The model is first pre-trained on simulation data to learn the general electromagnetic interference coupling mechanism, and then the model weights are fine-tuned on real data to adapt to the actual line characteristics. Use composite loss function for model optimization, combining regression loss and classification loss to achieve multi-objective optimization; Early stopping and cross-validation methods are used during training to evaluate model performance, prevent overfitting, and improve generalization capabilities. An online adaptive update mechanism is established. When the model prediction deviation exceeds the set threshold, the retraining process is triggered and the model parameters are updated using the newly collected data.

8. The method for fault location and electromagnetic interference resistance of distribution overhead lines according to claim 1, characterized in that: The steps for locating a distribution overhead line fault are: Use the trained electromagnetic disturbance propagation coupling model to analyze the real-time collected traveling wave signal and obtain the coupling characteristic matrix; The original traveling wave signal is subjected to interference suppression processing based on the coupling characteristic matrix, and the pure fault traveling wave component is separated from the mixed signal through the reverse decoupling algorithm; The purified fault traveling wave signal is used to calculate the traveling wave propagation time and reflection characteristics, and the fault location is determined in combination with the line propagation characteristics.

9. The traveling wave fault location and anti-electromagnetic interference system for distribution overhead lines is characterized by: The method for locating faults of distribution overhead lines with traveling waves and resisting electromagnetic interference according to any one of claims 1 to 8 comprises: The equivalent model building module is used to construct a broadband equivalent circuit model of the distribution overhead line based on the vector matching method, and is used to simulate the transmission fault signals of fault traveling waves and switching overvoltage; The fault signal analysis module uses a wavelet transform algorithm to perform multi-scale decomposition on the collected transmission fault signal, distinguish the signal type through time-frequency domain characteristics, extract the disturbance signal parameters, and calculate the traveling wave propagation characteristics for different fault types and switching conditions; A coupling model building module is used to combine a long short-term memory neural network and a convolutional neural network to build an electromagnetic disturbance propagation coupling model, taking the traveling wave propagation characteristics, disturbance signal parameters and line environment parameters as input and performing time-frequency domain analysis to obtain the coupling characteristics of the electromagnetic disturbance; The coupling model optimization module is used to perform numerical simulation using electromagnetic field simulation software, analyze the impact of line environment parameters on coupling characteristics based on simulation data, and compare and verify with field experimental data to optimize the parameters of the electromagnetic disturbance propagation coupling model; The fault location module is used to analyze the coupling characteristics of electromagnetic disturbances based on the optimized model, eliminate the impact of electromagnetic interference on traveling wave signals, and realize fault location of distribution overhead lines.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for locating faults of distribution overhead lines with traveling waves and resisting electromagnetic interference according to any one of claims 1 to 8 is implemented.

Images