A method for distinguishing lightning back strike fault from shielding failure fault of AC transmission line

Abstract

The present application relates to the technical field of measuring electrical variables, and particularly relates to a method for distinguishing lightning back strike fault and shielding failure fault of AC transmission line, comprising collecting lightning fault transient electrical signal and preprocessing, extracting time domain, frequency domain, time-frequency domain and waveform shape multi-dimensional feature parameters, and constructing multi-dimensional transient feature vector; simultaneously obtaining transmission line structure parameter information and standardizing processing; performing feature fusion on the multi-dimensional transient feature vector and structure parameter feature to generate fusion feature vector; constructing a differentiated identification model based on the fusion feature vector, and outputting fault type determination result. The present application can accurately distinguish the two types of faults and provide reliable basis for lightning protection optimization.

Description

Technical Field

[0001] This invention relates to the field of electrical variable measurement technology, and in particular to a method for distinguishing between lightning backflash faults and lightning strike faults in AC transmission lines. Background Technology

[0002] When AC transmission lines are struck by lightning, the interaction mechanism between the lightning discharge channel and the transmission line is fundamentally different. As a result, the backflash fault and the side-flash fault caused by this are completely different in terms of protection strategies. How to accurately identify the fault type is the key to ensuring the reliable operation of the line.

[0003] The transient electrical characteristics of backflashover faults and lightning strike faults are highly similar in waveform morphology, posing a serious challenge to traditional identification methods that rely on single waveform features. Common lightning fault identification methods generally use time-domain waveform parameter extraction, frequency-domain spectrum analysis, or single feature threshold comparison to determine the fault type. However, the waveform similarity leads to significant overlap in the use of single features to distinguish between the two types of faults, which limits the performance of existing methods in accurately identifying fault types and makes it difficult to provide reliable fault type identification. At the same time, protective measures are misplaced due to identification errors, making it impossible to achieve targeted and differentiated lightning protection. Summary of the Invention

[0004] The purpose of this invention is to provide a method for distinguishing between lightning backflash faults and lightning strike faults in AC transmission lines, so as to solve the technical problems mentioned in the background art.

[0005] This invention provides a method for distinguishing between lightning backflashover faults and lightning strike faults in AC transmission lines, comprising: The transient electrical signal data of a transmission line during a lightning fault are collected, and the transient electrical signal data is preprocessed to eliminate noise interference and signal distortion, so as to obtain a preprocessed transient electrical signal that meets the feature extraction requirements. Multiple transient feature parameters of various dimensions are extracted from the preprocessed transient electrical signal to construct a multi-dimensional transient feature vector, which includes time-domain feature parameters, frequency-domain feature parameters, time-frequency-domain feature parameters, and waveform morphology feature parameters. Obtain the line structure parameter information of the transmission line, and standardize the line structure parameter information; The multi-dimensional transient feature vector is fused with the standardized line structure parameter information to generate a fused feature vector. The feature fusion process is implemented through a weight allocation mechanism. A differentiated identification model is constructed based on the fused feature vector, and the differentiated identification model is used to classify and identify the current lightning fault, and output the fault type determination result.

[0006] In some embodiments, the acquisition of transient electrical signal data when a lightning fault occurs in a transmission line includes at least: synchronously acquiring three-phase voltage signals and three-phase current signals at the moment the fault occurs through voltage transformers and current transformers installed on each phase conductor of the transmission line; the sampling frequency of the voltage transformers and current transformers is not less than 10 million sampling points per second; and synchronously calibrating the acquired raw voltage signals and raw current signals to eliminate signal time offset caused by differences in sensor installation positions.

[0007] In some embodiments, the method further includes: calculating the instantaneous power waveform during the fault based on the acquired voltage and current signals, and extracting the overvoltage multiple, overcurrent multiple, and power change characteristics from the instantaneous power waveform.

[0008] In some embodiments, extracting multiple-dimensional transient feature parameters from the preprocessed transient electrical signal includes at least: extracting time-domain feature parameters, specifically: Calculate the peak factor, pulse factor, waveform factor, peak time, rise time, fall time, and signal energy value of the preprocessed transient electrical signal; the peak factor is the ratio of the signal peak value to the effective value, the pulse factor is the ratio of the signal peak value to the rectified average value, and the waveform factor is the ratio of the effective value to the rectified average value.

[0009] In some embodiments, the method further includes: extracting frequency domain feature parameters, specifically: The preprocessed transient electrical signal is subjected to a fast Fourier transform to obtain the signal spectrum; Calculate the fundamental amplitude, the amplitude of each harmonic, and the total harmonic distortion rate based on the signal spectrum; The signal spectrum is divided into low-frequency, mid-frequency, and high-frequency bands, and the energy proportion of each band is calculated.

[0010] In some embodiments, the method further includes: extracting time-frequency domain feature parameters, specifically: Wavelet transform is performed on the preprocessed transient electrical signal to extract the energy distribution characteristics and coefficient modulus maxima distribution characteristics of each level of wavelet decomposition. The Hilbert-Huang transform is used to extract the intrinsic mode function and instantaneous frequency characteristics of the signal.

[0011] In some embodiments, the method further includes: extracting waveform morphology feature parameters, specifically: calculating the wavefront steepness, zero-crossing rate, waveform asymmetry, and polarity reversal characteristics of the preprocessed transient electrical signal waveform; wherein the wavefront steepness is the ratio of the signal peak value to the rise time.

[0012] In some embodiments, obtaining the line structure parameter information of the transmission line includes at least: The tower type, number of insulator strings, length of insulator strings, conductor type, conductor arrangement, and soil resistivity of the line corridor are obtained from the transmission line parameter database. Historical lightning activity data of the line corridor is obtained through a lightning location system, including the probability density function of lightning current amplitude distribution, daily lightning distribution density, and the statistical characteristics of the probability of lightning striking the conductor and ground wire.

[0013] In some embodiments, the method further includes: obtaining the measured value of the grounding resistance of the line tower through on-site testing, and correcting the grounding resistance value in combination with seasonal variation factors.

[0014] In some embodiments, the feature fusion processing of the multi-dimensional transient feature vector and the standardized line structure parameter information includes at least: The numerical range of each feature component in the multidimensional transient feature vector is mapped to the interval between zero and one. Principal component analysis was performed on the standardized line structure parameter information to extract principal component eigenvectors and retain principal components with a cumulative contribution rate of over 85%. Based on a combined weighting method that combines the analytic hierarchy process (AHP) and the entropy weighting method, the comprehensive weight coefficient of each fusion feature dimension is calculated. The multi-dimensional transient feature vector after mapping is concatenated with the principal component feature vector by weighting according to the comprehensive weight coefficient to generate the fused feature vector.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By integrating multi-dimensional transient features and line structure parameters to construct a differentiated identification model, the technical bottleneck of traditional methods relying on single waveform features, which makes it difficult to distinguish between backflash faults and backflash faults, is effectively overcome. 2. The extracted multi-dimensional transient features comprehensively characterize the electrical characteristics of lightning faults from four complementary perspectives: time domain, frequency domain, time-frequency domain, and waveform morphology, which can capture the subtle differences between the two types of faults. At the same time, the introduced line structure parameters reflect the lightning protection attributes of the transmission line itself, enabling the identification model to make comprehensive judgments in combination with the specific operating environment of the line. 3. The feature fusion processing adopted rationally integrates the distinguishing abilities of transient features and structural features through a weight allocation mechanism, so that the fused feature vector has a stronger ability to represent fault types; 4. The constructed differential recognition model integrates three different classifier architectures: support vector machine, random forest, and convolutional neural network. By combining the advantages of each classifier through a weighted voting mechanism, the reliability and stability of the classification results are improved. 5. Provides reliable fault type identification basis for lightning protection optimization, enabling operation and maintenance personnel to take targeted and differentiated lightning protection measures based on the identification results, avoiding protection misalignment, and improving the reliability of AC transmission line operation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the method for distinguishing between lightning backflash faults and lightning strike faults in AC transmission lines according to the present invention. Figure 2 This is a flowchart of the multi-dimensional transient feature vector and standardized line structure parameter information feature fusion processing of the present invention. Detailed Implementation

[0018] The following will be based on embodiments of the present invention. Figures 1-2 The technical solutions in the embodiments of the present invention will be clearly and completely described together. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] To better understand this invention, some definitions are provided below: Hilbert Huang Transform: An adaptive time-frequency analysis method specifically designed for nonlinear and non-stationary signals. It is not a fixed-basis transform, but a data-driven algorithm process.

[0020] Example 1 First, transient electrical signal data of the transmission line when lightning fault occurs are collected, and the transient electrical signal data are preprocessed to eliminate noise interference and signal distortion, so as to obtain preprocessed transient electrical signals that meet the feature extraction requirements. The transient electrical signals include voltage and current waveform data at the moment the fault occurs.

[0021] The three-phase voltage and current signals at the moment of a fault occurrence are synchronously acquired by voltage transformers and current transformers installed on each phase conductor of the transmission line. The sampling frequency of the voltage and current transformers is set to 10 million sampling points per second to ensure that all high-frequency components of the lightning transient signal are captured. The three-phase voltage signals are labeled Ua, Ub, and Uc, and the three-phase current signals are labeled Ia, Ib, and Ic.

[0022] The acquired raw voltage and current signals undergo synchronous calibration to eliminate signal time offsets caused by differences in sensor installation positions. This synchronous calibration process includes three stages: timestamp calibration, phase calibration, and amplitude calibration. Timestamp calibration ensures that the sampling data from all channels has a unified time reference, achieving nanosecond-level resolution. Phase calibration compensates for phase differences between voltage and current transformers, ensuring accurate phase relationships between voltage and current signals. Amplitude calibration normalizes the signals based on the transformer ratio and phase characteristics, resulting in standardized voltage and current waveform data.

[0023] Furthermore, the instantaneous power waveform during the fault is calculated based on the acquired voltage and current signals. The instantaneous power is obtained by multiplying the voltage and current signals, i.e., P(t) = u(t) × i(t). During the calculation, according to the power calculation principle of a three-phase circuit, the instantaneous power of each phase and the total active power of the three phases are calculated separately. The overvoltage multiple, overcurrent multiple, and power surge characteristics are extracted from the instantaneous power waveform. The overvoltage multiple is defined as the ratio of the peak voltage during the fault to the rated voltage, and the overcurrent multiple is defined as the ratio of the peak current during the fault to the rated current. Power surge characteristics include parameters such as the power rise rate, power fluctuation amplitude, and power recovery time.

[0024] Furthermore, for the acquired raw transient electrical signal, denoising and distortion correction are first performed. Wavelet threshold denoising is used to decompose the raw signal and reconstruct the denoised signal. During wavelet threshold denoising, appropriate wavelet basis functions and decomposition levels are selected to decompose the signal into high-frequency and low-frequency coefficients. Soft thresholding is applied to the high-frequency coefficients according to threshold rules, and then the denoised signal is reconstructed. Distortion correction includes baseline drift correction and zero-point drift correction. A high-pass filter is used to filter out baseline drift components below a set frequency, ensuring the signal baseline remains stable near zero potential.

[0025] Next, multiple transient feature parameters are extracted from the preprocessed transient electrical signal to construct a multi-dimensional transient feature vector, which includes time-domain feature parameters, frequency-domain feature parameters, time-frequency-domain feature parameters, and waveform morphology feature parameters.

[0026] Time-domain characteristic parameters are used to characterize the peak value, rise time, duration, and energy distribution characteristics of fault transient signals. Frequency-domain characteristic parameters are used to characterize the energy proportion and harmonic distribution characteristics of fault transient signals in different frequency bands. Time-frequency domain characteristic parameters are used to characterize the joint distribution characteristics of fault transient signals on the time and frequency axes. Waveform morphology characteristic parameters are used to characterize the wavefront steepness, zero-crossing characteristics, and polarity characteristics of fault transient signals.

[0027] Furthermore, the extraction method for time-domain feature parameters employs multiple statistical calculation methods. These can be understood as follows: Calculating the peak factor of the preprocessed transient electrical signal (the ratio of the signal peak value to the effective value, used to characterize the signal's impulse characteristics); calculating the impulse factor (the ratio of the signal peak value to the rectified average value, used to characterize the signal's pulse intensity); calculating the waveform factor (the ratio of the effective value to the rectified average value, used to characterize the integrity of the signal waveform); calculating the peak time (the time interval from the signal's initial point to its peak); calculating the rise time (the time it takes for the signal amplitude to increase from 10% to 90%); calculating the fall time (the time it takes for the signal amplitude to decrease from 90% to 10%); and calculating the signal energy value (integrating the square of the signal amplitude to obtain the total signal energy).

[0028] Furthermore, the frequency domain feature parameter extraction method first performs a Fast Fourier Transform (FFT) on the preprocessed transient electrical signal to obtain the signal spectrum. The number of points in the FFT matches the length of the original signal to ensure that the spectral resolution meets the analysis requirements. The fundamental amplitude is calculated based on the signal spectrum, i.e., the amplitude corresponding to the power frequency component. The amplitudes of each harmonic are calculated, including the amplitudes of the second, third, and fourth harmonics. The total harmonic distortion (THD) is calculated, defined as the ratio of the root mean square (RMS) value of each harmonic amplitude to the fundamental amplitude. The signal spectrum is divided into three frequency bands: low-frequency, mid-frequency, and high-frequency. The low-frequency band is defined as the frequency range from zero to 1000 Hz, the mid-frequency band as the frequency range from 1000 Hz to 100,000 Hz, and the high-frequency band as the frequency range from 100,000 Hz to half the sampling frequency. The energy proportion of each frequency band is calculated to obtain the frequency domain energy distribution feature vector.

[0029] Furthermore, the extraction method for time-frequency domain feature parameters utilizes two time-frequency analysis tools: wavelet transform and Hilbert-Huang transform. Wavelet transform is applied to the preprocessed transient electrical signal, and a multi-scale decomposition method is used to decompose the signal into components at different frequency levels. The number of wavelet decomposition levels is determined based on the signal characteristics and required analysis accuracy, typically five to seven levels. The energy distribution characteristics of each level of wavelet decomposition are extracted, i.e., the proportion of energy corresponding to each decomposition level to the total energy. The distribution characteristics of the modulus maxima of the wavelet coefficients are extracted, i.e., the location and amplitude of local maxima points of the wavelet coefficient modulus, which correspond to the abrupt changes in the signal. The intrinsic mode functions (IMFs) of the signal are extracted using Hilbert-Huang transform, and the signal is decomposed into several IMFs through empirical mode decomposition, each IMF representing an intrinsic oscillation mode. The instantaneous frequency characteristics are extracted, i.e., the variation of the instantaneous frequency of the IMFs over time.

[0030] Furthermore, the method for extracting waveform morphology feature parameters focuses on the geometric morphology of the signal. It calculates the wavefront steepness of the preprocessed transient electrical signal waveform, which is the ratio of the signal peak value to the rise time, characterizing the sharpness of the signal wavefront. It calculates the zero-crossing rate, the number of times the signal crosses zero potential per unit time, reflecting the signal's oscillation frequency. It calculates waveform asymmetry, defined as the degree of asymmetry between the signal's rise and fall processes, obtained by comparing the differences in rise and fall times. Finally, it extracts polarity reversal features, detecting whether the signal undergoes polarity reversal and the number and timing of such reversals.

[0031] Next, the line structure parameter information of the transmission line is obtained and the line structure parameter information is standardized.

[0032] The line structure parameters include the line tower grounding resistance, insulator string flashover characteristics, lightning activity intensity distribution along the line corridor, line altitude, and line shielding protection method.

[0033] The transmission line parameter database is used to obtain information such as tower type, number of insulator discs, insulator string length, conductor type, conductor arrangement, and soil resistivity of the transmission line corridor. Tower type includes information such as tower height, crossarm length, and grounding down conductor location; the number of insulator discs determines the insulation strength of the insulator string, while the insulator string length affects flashover voltage characteristics; conductor type includes parameters such as conductor cross-sectional area, material, and rated current; conductor arrangement includes different layouts such as horizontal, vertical, and triangular arrangements; soil resistivity is a key parameter affecting grounding resistance, and the resistivity varies significantly among different soil types.

[0034] Furthermore, historical lightning activity data of the power line corridor is obtained through a lightning location system, including the probability density function of lightning current amplitude distribution, daily lightning distribution density, and the statistical characteristics of lightning strikes on conductors and ground wires. The lightning location system employs a network of multiple detection stations to determine the location and time of lightning occurrences by measuring the electromagnetic radiation signals generated by lightning discharges. The probability density function of lightning current amplitude distribution describes the probability distribution of different amplitude lightning currents; the daily lightning distribution density represents the number of days with lightning activity each year, reflecting the frequency of lightning activity along the power line corridor; and the statistical characteristics of lightning strikes on conductors and ground wires are calculated based on historical lightning location data and power line geometric parameters.

[0035] Furthermore, the grounding resistance of the transmission towers was obtained through on-site testing. The on-site testing employed a three-electrode method using a grounding resistance meter, with the current and voltage electrodes positioned appropriately to ensure accuracy. The grounding resistance values ​​were then corrected for seasonal variations, including soil moisture, temperature, and frost depth. Significant differences in soil resistivity exist across seasons, necessitating seasonal correction to obtain representative values.

[0036] Then, the multi-dimensional transient feature vectors are fused with the standardized line structure parameter information to generate a fused feature vector. The feature fusion process is implemented through a weight allocation mechanism. The weight allocation is determined based on the ability of each feature dimension to distinguish between backflashover faults and bypass faults.

[0037] In detail, each feature component in the multi-dimensional transient feature vector is normalized, mapping the numerical range of each feature component to the interval between zero and one. The normalization process ensures that features of different dimensions and orders of magnitude are comparable, avoiding feature weight deviations caused by numerical differences.

[0038] Furthermore, principal component analysis is performed on the standardized line structure parameter information to extract principal component eigenvectors. Principal component analysis is a dimensionality reduction method that projects the original feature space into a new orthogonal coordinate system through linear transformation. The resulting principal components are arranged in descending order of variance contribution rate, and principal components with a cumulative contribution rate of more than 85% are retained to ensure that the dimensionality-reduced feature space retains most of the original information.

[0039] Furthermore, a combined weighting method based on the analytic hierarchy process (AHP) and entropy weighting is used to calculate the comprehensive weight coefficients for each fused feature dimension. The AHP establishes a hierarchical structure model to analyze the relative importance relationships between factors, obtaining the subjective weights of each factor. The entropy weighting method determines the objective weights based on the information entropy values ​​of each factor; the lower the information entropy, the greater the variability and the stronger the discriminative ability of that factor. The subjective weights obtained from the AHP and the objective weights obtained from the entropy weighting method are combined to obtain the comprehensive weight coefficients. The combination method uses a linear weighting approach. The overall weight is calculated as α × subjective weight + β × objective weight, where α and β are preset combination coefficients.

[0040] The normalized multidimensional transient feature vector and the principal component feature vector are weighted and concatenated according to the comprehensive weight coefficient to generate a fused feature vector. In the weighted concatenation process, the feature values ​​at corresponding positions of the two feature vectors are weighted and fused according to the corresponding comprehensive weight coefficient to form a new fused feature vector. The dimension of the fused feature vector is jointly determined by the dimension of the original feature vector and the dimension of the principal component feature vector, and its length is equal to the sum of the dimension of the multidimensional transient feature vector and the dimension of the principal component feature vector.

[0041] Finally, a differentiated identification model is constructed based on the fused feature vectors, and this model is used to classify and identify the current lightning fault, outputting the fault type determination result. The differentiated identification model is a classification model trained using machine learning algorithms, which includes a feature mapping layer, a feature selection layer, and a classification decision layer.

[0042] An initial classification model based on Support Vector Machines (SVMs) is constructed. SVMs are classification algorithms based on statistical learning theory, and their core idea is to find the optimal classifying hyperplane in the feature space. SVMs use radial basis functions (RBFs) as kernel functions. RBFs can map the original feature space to a higher-dimensional space, making originally linearly inseparable data linearly separable. The kernel function parameters and penalty factor are determined using a grid search method. The kernel function parameters control the width of the RBFs, and the penalty factor controls the degree of punishment for misclassified samples. The grid search performs an exhaustive search within a preset parameter range, selecting the parameter combination that maximizes the cross-validation accuracy.

[0043] A random forest-based auxiliary classification model is constructed. Random forest is an ensemble learning method based on decision trees, which classifies data by constructing multiple decision trees and combining their outputs. The number of decision trees in the random forest is set to 100 to 300; the more decision trees, the stronger the model's stability and generalization ability. During training, each decision tree uses random sampling to select samples and features, reducing the risk of overfitting. The maximum depth is set to 10 to 20 layers to limit the complexity of the decision trees and prevent overfitting.

[0044] A deep classification model based on convolutional neural networks (CNNs) is constructed. CNNs are a type of deep learning model that excels at processing data with a grid structure. The deep classification model includes an input layer, several convolutional layers, several pooling layers, a fully connected layer, and an output layer. The input layer receives a fused feature vector as input. The convolutional layers perform convolution operations with the input data using convolution kernels to extract local features. The pooling layers downsample the convolutional output, reducing the feature dimensionality and enhancing the model's translation invariance. The fully connected layer flattens the previous feature maps into a one-dimensional vector for feature integration. The output layer uses the Softmax function to output the probability values ​​for each category.

[0045] The initial classification model, auxiliary classification model, and deep classification model were trained using a pre-defined fault sample dataset to obtain their respective classification performance metrics. The fault sample dataset included a large number of labeled lightning fault samples, each containing a fused feature vector and a corresponding fault type label. Cross-validation was used to evaluate model performance during training, and classification performance metrics included accuracy, precision, recall, F1 score, and area under the ROC curve.

[0046] Based on classification performance metrics, a weighted voting mechanism is used to integrate the outputs of three classification models to generate the final differentiated identification model. The weighted voting mechanism assigns voting weights according to the performance metrics of each classification model, with higher-performing models receiving higher voting weights. Each classification model outputs a category prediction result and corresponding confidence score for the input sample. The outputs of the three models are then weighted and fused to obtain the final classification decision result.

[0047] Differential recognition models are used to classify and identify current lightning faults. The fused feature vector corresponding to the current lightning fault is input into the differential recognition model, and the category probability values ​​output by each classification model are obtained. The category probability value represents the likelihood that the input sample belongs to each fault category. After inputting the fused feature vector corresponding to the current lightning fault into the differential recognition model, the Support Vector Machine model outputs two probability values: one for a counter-attack fault and the other for a bypass fault. Similarly, the Random Forest model and the Convolutional Neural Network model also output two probability values.

[0048] A weighted average method is used to calculate the weighted average of the category probabilities of each classification model, and the category with the largest weighted average is selected as the predicted category. Specifically, Let the voting weights of the three models be w1, w2, and w3, and the counterattack failure probabilities output by each model be p1, p2, and p3, respectively. Then the weighted average counterattack failure probability is: (w1×p1+w2×p2+w3×p3) / (w1+w2+w3).

[0049] After comparing the probability values ​​of the two types of faults with a weighted average, the category with the larger probability value is used as the predicted category.

[0050] Furthermore, a Bayesian posterior probability correction mechanism is introduced, which combines the prior probability to correct the predicted category and output the final fault type determination result. The Bayesian posterior probability correction is based on Bayes' theorem, combining the prior probability with the likelihood probability output by the model to obtain the posterior probability. The prior probability is obtained from historical statistical data and represents the probability of each fault category occurring in the absence of any observational information. When there is a significant difference between the model output and the prior probability, the posterior probability correction mechanism adjusts the model output to obtain a more reasonable classification result.

[0051] Example 2 The following explanation uses a 500 kV AC transmission line in a certain region as an example.

[0052] The line is approximately 150 kilometers long, traversing both mountainous and plain terrains, with a total of about 300 towers and glass insulator strings.

[0053] When a lightning fault occurs on the line, voltage transformers and current transformers installed at key towers synchronously collect the transient fault signals, with a sampling frequency set to 10 million sampling points per second. The collected three-phase voltage and current signals are transmitted to the background monitoring system via fiber optic communication. The background system performs synchronous calibration and noise reduction preprocessing on the raw signals, filtering out power frequency interference and random noise to obtain clean transient fault waveform data.

[0054] The backend system extracts multi-dimensional feature parameters from the preprocessed transient electrical signal. Time-domain feature extraction yields parameters such as peak factor, impulse factor, waveform factor, peak time, rise time, fall time, and signal energy. Frequency-domain feature extraction yields parameters such as fundamental amplitude, harmonic amplitudes, total harmonic distortion (THD), low-frequency energy proportion, mid-frequency energy proportion, and high-frequency energy proportion. Time-frequency domain feature extraction yields parameters such as energy distribution at each wavelet decomposition level, intrinsic mode functions, and instantaneous frequency. Waveform morphology feature extraction yields parameters such as wavefront steepness, zero-crossing rate, waveform asymmetry, and polarity reversal characteristics. These feature parameters constitute a 180-dimensional multi-dimensional transient feature vector.

[0055] Simultaneously, the system retrieves the structural parameters of the transmission line from the transmission line parameter database, including tower type, number of insulator discs, conductor type, conductor arrangement, and soil resistivity. Historical lightning data of the line corridor is obtained from the lightning location system, including lightning current amplitude distribution, daily lightning distribution, and probability of lightning striking the conductor. The grounding resistance values ​​of each tower are measured on-site and seasonally corrected. After standardization, these structural parameters are subjected to principal component analysis to extract principal component eigenvectors.

[0056] The 128-dimensional transient feature vector and the principal component feature vector are weighted and concatenated to generate a fused feature vector. The fused feature vector has 150 dimensions, with 128 dimensions for transient features and 22 dimensions for principal component features. The fused feature vector is input into a differential recognition model for classification. The model outputs the probabilities of counterattack failure and bypass failure. After correction based on the Bayesian posterior probability, the final failure type determination result is output.

[0057] Based on the fault type identification results, maintenance personnel take targeted measures. If the fault is determined to be a backflash fault, the focus is on checking the operating status of the tower grounding device and insulator string, and taking protective measures such as reducing grounding resistance and strengthening insulation. If the fault is determined to be a backflash fault, the focus is on checking the line shielding measures and the lightning protection angle, and taking protective measures such as optimizing the shielding design and adding coupling ground wires.

[0058] By accurately identifying fault types, differentiated configuration of lightning protection measures can be achieved, avoiding misalignment of protection and improving the reliability of line operation.

[0059] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0060] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment includes only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for distinguishing between lightning backflashover faults and lightning strike faults in AC transmission lines, characterized in that, include: The transient electrical signal data of a transmission line during a lightning fault are collected, and the transient electrical signal data is preprocessed to eliminate noise interference and signal distortion, so as to obtain a preprocessed transient electrical signal that meets the feature extraction requirements. Multiple transient feature parameters of various dimensions are extracted from the preprocessed transient electrical signal to construct a multi-dimensional transient feature vector, which includes time-domain feature parameters, frequency-domain feature parameters, time-frequency-domain feature parameters, and waveform morphology feature parameters. Obtain the line structure parameter information of the transmission line, and standardize the line structure parameter information; The multi-dimensional transient feature vector is fused with the standardized line structure parameter information to generate a fused feature vector. The feature fusion process is implemented through a weight allocation mechanism. A differentiated identification model is constructed based on the fused feature vector, and the differentiated identification model is used to classify and identify the current lightning fault, and output the fault type determination result.

2. The method according to claim 1, characterized in that, The acquisition of transient electrical signal data during lightning faults in transmission lines includes at least the following: synchronous acquisition of three-phase voltage and three-phase current signals at the moment of fault occurrence via voltage transformers and current transformers installed on each phase conductor of the transmission line; the sampling frequency of the voltage transformers and current transformers is not less than 10 million sampling points per second; and synchronous calibration processing of the acquired raw voltage and raw current signals to eliminate signal time offset caused by differences in sensor installation positions.

3. The method according to claim 2, characterized in that, It also includes: calculating the instantaneous power waveform during the fault based on the collected voltage and current signals, and extracting the overvoltage multiple, overcurrent multiple, and power change characteristics from the instantaneous power waveform.

4. The method according to claim 1, characterized in that, The extraction of multi-dimensional transient feature parameters from the preprocessed transient electrical signal includes at least: extracting time-domain feature parameters, specifically: Calculate the peak factor, pulse factor, waveform factor, peak time, rise time, fall time, and signal energy value of the preprocessed transient electrical signal; the peak factor is the ratio of the signal peak value to the effective value, the pulse factor is the ratio of the signal peak value to the rectified average value, and the waveform factor is the ratio of the effective value to the rectified average value.

5. The method according to claim 1, characterized in that, Also includes: Extracting frequency domain feature parameters, specifically: The preprocessed transient electrical signal is subjected to a fast Fourier transform to obtain the signal spectrum; Calculate the fundamental amplitude, the amplitude of each harmonic, and the total harmonic distortion rate based on the signal spectrum; The signal spectrum is divided into low-frequency, mid-frequency, and high-frequency bands, and the energy proportion of each band is calculated.

6. The method according to claim 1, characterized in that, Also includes: Extracting time-frequency domain feature parameters, specifically: Wavelet transform is performed on the preprocessed transient electrical signal to extract the energy distribution characteristics and coefficient modulus maxima distribution characteristics of each level of wavelet decomposition. The Hilbert-Huang transform is used to extract the intrinsic mode function and instantaneous frequency characteristics of the signal.

7. The method according to claim 1, characterized in that, Also includes: The waveform morphology feature parameters are extracted, specifically: the wavefront steepness, zero-crossing rate, waveform asymmetry, and polarity reversal characteristics of the preprocessed transient electrical signal waveform are calculated; the wavefront steepness is the ratio of the signal peak value to the rise time.

8. The method according to claim 1, characterized in that, The acquisition of transmission line structure parameter information includes at least: The tower type, number of insulator strings, length of insulator strings, conductor type, conductor arrangement, and soil resistivity of the line corridor are obtained from the transmission line parameter database. Historical lightning activity data of the line corridor is obtained through a lightning location system, including the probability density function of lightning current amplitude distribution, daily lightning distribution density, and the statistical characteristics of the probability of lightning striking the conductor and ground wire.

9. The method according to claim 8, characterized in that, Also includes: The measured values ​​of grounding resistance of the line towers were obtained through on-site testing, and the grounding resistance values ​​were corrected by taking into account seasonal variation factors.

10. The method according to claim 1, characterized in that, The feature fusion process of combining the multi-dimensional transient feature vector with the standardized line structure parameter information includes at least the following: The numerical range of each feature component in the multidimensional transient feature vector is mapped to the interval between zero and one. Principal component analysis was performed on the standardized line structure parameter information to extract principal component eigenvectors and retain principal components with a cumulative contribution rate of over 85%. Based on a combined weighting method that combines the analytic hierarchy process (AHP) and the entropy weighting method, the comprehensive weight coefficient of each fusion feature dimension is calculated. The multi-dimensional transient feature vector after mapping is concatenated with the principal component feature vector by weighting according to the comprehensive weight coefficient to generate the fused feature vector.

Images