Cable fault positioning method and system based on neural network learning

Abstract

The invention relates to the field of electrical system fault detection, and provides a cable fault positioning method and system based on neural network learning. The method comprises the following steps: acquiring a cable traveling wave signal and preprocessing the cable traveling wave signal to obtain a preprocessed traveling wave signal; performing high-speed parallel acquisition on the preprocessed traveling wave signal to obtain multi-channel digital traveling wave data; performing multi-resolution decomposition on the multi-channel digital traveling wave data through a fast wavelet transform algorithm, and performing feature extraction on a wavelet coefficient obtained through decomposition to obtain a fault feature vector; and carrying out classification identification on the fault feature vector through a multilayer perceptron neural network, and carrying out fault position calculation on a classification result to obtain a fault positioning result. According to the invention, the cable fault positioning precision and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical system fault detection, and in particular to a cable fault locating method and system based on neural network learning. Background Art

[0002] As a vital component of modern power systems, the operating status of power cables directly impacts the safety and stability of the power grid. Traditional cable fault detection technologies primarily include impedance, bridge, and pulse reflection methods, which are relatively mature for detecting cable core faults. In recent years, traveling wave fault location technology has gained widespread application in power system fault detection due to its high positioning accuracy and fast response speed. Existing traveling wave location systems typically employ two-terminal or multi-terminal measurement methods to determine the fault location by analyzing the propagation characteristics of the traveling wave signal generated by the fault in the cable.

[0003] However, in the comprehensive detection of existing cable faults, existing signal processing methods are unable to meet the requirements of filtering under strong electromagnetic interference conditions, resulting in a serious decline in signal quality and affecting the accuracy of fault identification. Secondly, traditional fault classification methods mainly rely on manually set judgment rules and thresholds, lack adaptive learning capabilities, and have low recognition accuracy when facing new fault modes. In addition, existing positioning algorithms fail to fully utilize the multi-dimensional feature information of traveling wave signals, resulting in insufficient positioning accuracy in complex fault scenarios, especially when multiple faults occur simultaneously. Misjudgment is prone to occur. Summary of the Invention

[0004] The present invention provides a cable fault location method and system based on neural network learning, which are used to solve the defects of the prior art.

[0005] The present invention provides a cable fault location method based on neural network learning, comprising: S1: collecting a cable traveling wave signal and preprocessing the cable traveling wave signal to obtain a preprocessed traveling wave signal; S2: performing high-speed parallel acquisition on the pre-processed traveling wave signal to obtain multi-channel digital traveling wave data; S3: performing multi-resolution decomposition on the multi-channel digitized traveling wave data using a fast wavelet transform algorithm, and performing feature extraction on the wavelet coefficients obtained by the decomposition to obtain a fault feature vector; S4: Classify and identify the fault feature vector through a multi-layer perceptron neural network, and calculate the fault location based on the classification result to obtain a fault location result.

[0006] According to a cable fault location method based on neural network learning provided by the present invention, step S1 further includes: S11: collect cable traveling wave signals; S12: performing multi-band filtering on the cable traveling wave signal through a high-pass filter and a low-pass filter to obtain a filtered traveling wave signal; S13: performing noise suppression on the filtered traveling wave signal by using an LMS adaptive filtering algorithm to obtain a preprocessed traveling wave signal.

[0007] According to a cable fault location method based on neural network learning provided by the present invention, step S12 further includes: S121: Input the cable traveling wave signal into a high-pass filter to filter out baseband interference and obtain a high-frequency traveling wave component; S122: Input the high-frequency traveling wave component into a low-pass filter for anti-aliasing filtering to obtain a band-limited filtered traveling wave signal.

[0008] According to a cable fault location method based on neural network learning provided by the present invention, step S2 further includes: S21: Distributing the pre-processed traveling wave signal to multiple LVDS differential output ADC chips for synchronous digital conversion to obtain a parallel digital sequence; S22: Perform clock synchronization calibration on the parallel digital sequence to obtain the calibration traveling wave data; S23: The calibration traveling wave data is stored in a high-speed cache, and valid segments of the cached data are extracted according to a preset trigger condition to obtain multi-channel digitized traveling wave data related to the fault event.

[0009] According to a cable fault location method based on neural network learning provided by the present invention, step S3 further includes: S31: Selecting a Daubechies wavelet basis, performing multi-resolution and multi-scale decomposition on the multi-channel digitized traveling wave data, and obtaining wavelet coefficients corresponding to multiple frequency bands; S32: performing adaptive threshold denoising on the wavelet coefficients of multiple frequency bands to obtain a denoised wavelet coefficient matrix; S33: Perform feature extraction on the denoised wavelet coefficient matrix to obtain a fault feature vector.

[0010] According to a cable fault location method based on neural network learning provided by the present invention, step S33 further includes: S331: Calculating the time-domain statistical features of the denoised wavelet coefficient matrix to extract the first-order features of the denoised wavelet coefficient matrix and obtain a time-domain feature set; S332: Calculating the frequency domain energy distribution of the denoised wavelet coefficient matrix to extract the second-order features of the denoised wavelet coefficient matrix and obtain a frequency domain feature set; S333: Fusing the time domain feature set and the frequency domain feature set to obtain a fault feature vector.

[0011] According to a cable fault location method based on neural network learning provided by the present invention, step S4 further includes: S41: Inputting the fault feature vector into a multilayer perceptron neural network to obtain a fault type probability distribution; S42: Determine the dominant fault type according to the fault type probability distribution and obtain a fault type identifier; S43: Calculating the arrival time difference of the traveling waves at both ends based on the multi-channel digitized traveling wave data to perform time difference measurement and obtain the traveling wave propagation time difference; S44: Using a double-end positioning method, distance calculation is performed according to the traveling wave propagation time difference to obtain the fault positioning result.

[0012] According to a cable fault location method based on neural network learning provided by the present invention, step S41 further includes: S411: Perform forward propagation calculation on the fault feature vector through a multilayer perceptron neural network to obtain a hidden layer activation output; S412: Performing Softmax normalization processing on the hidden layer activation output to obtain a fault type probability distribution.

[0013] According to a cable fault location method based on neural network learning provided by the present invention, the expression of the fault location result in step S44 is: ; in, To calculate the fault location result, is the traveling wave propagation time difference, is the base time difference, is the propagation speed of the traveling wave in the cable medium, is the collective structure correction coefficient, is the frequency-dependent correction factor, is the cable length, To monitor the vertical distance between the terminal and the fault point, is the traveling wave frequency, It is the horizontal distance between the monitoring terminal and the fault point.

[0014] The present invention also provides a cable fault location system based on neural network learning, comprising: Preprocessing module: used for collecting cable traveling wave signals and preprocessing the cable traveling wave signals to obtain preprocessed traveling wave signals; An acquisition module, configured as a multi-channel LVDS differential output ADC chip, for performing high-speed parallel acquisition of the pre-processed traveling wave signal to obtain multi-channel digitized traveling wave data; Extraction module: used to perform multi-resolution decomposition on the multi-channel digitized traveling wave data through a fast wavelet transform algorithm, and perform feature extraction on the wavelet coefficients obtained by the decomposition to obtain a fault feature vector; an identification module, wherein the identification module is configured as a multi-layer perceptron neural network and is used to classify and identify the fault feature vector to obtain a classification result; Positioning module: used to calculate the fault location based on the classification result to obtain the fault location result.

[0015] The present invention provides a cable fault location method and system based on neural network learning. Through data preprocessing with multi-mechanism fusion, it can significantly improve signal quality in complex electromagnetic interference environments. The LMS adaptive filtering algorithm has the intelligent characteristic of dynamically adjusting filtering parameters according to noise statistical characteristics, and can automatically adapt to interference patterns under different working conditions. Compared with traditional fixed parameter filtering methods, the signal-to-noise ratio is improved, effectively solving the technical problem of the simultaneous existence of 50Hz fundamental frequency interference and high-frequency noise in the power system. Secondly, the present invention realizes the synergistic effect of multi-channel synchronization technology through the LVDS differential output ADC chip, which not only ensures the complete capture of weak fault signals, but more importantly, realizes the time synchronization acquisition of the main core line and sheath loop signals, providing a high-precision time reference for the subsequent two-end traveling wave time difference positioning algorithm, thereby improving the fault location accuracy. Compared with the traditional single-channel acquisition method, the positioning accuracy is also improved. The present invention also applies the fast wavelet transform algorithm in multi-resolution decomposition processing, giving full play to the advantages of wavelet transform in time-frequency domain analysis. By decomposing the traveling wave signal into wavelet coefficients of different frequency bands, it can not only effectively separate the transient state in the fault signal, but also effectively separate the transient state in the fault signal. The components and steady-state components can better capture the time-frequency characteristic change law at the moment of fault occurrence, provide a rich source of characteristic information for the statistical analysis algorithm, and greatly improve the dimension and characterization ability of the fault feature vector; in addition, the present invention uses the deep learning ability of the multi-layer perceptron neural network in the fault classification and identification processing, and establishes a multi-layer nonlinear mapping relationship to automatically learn and extract the deep characteristic laws of the fault mode. Compared with the traditional rule-based expert system, the fault classification accuracy is improved. Especially when facing new fault modes, the generalization ability of the neural network enables the system to have the intelligent characteristics of continuous learning and self-optimization, which effectively solves the technical limitations of the traditional method that relies on manual experience to set the discrimination threshold; finally, the double-end traveling wave time difference positioning algorithm is applied to realize the organic unity of fault type discrimination and position positioning. By using the fault type probability distribution output by the neural network as a weighting coefficient to comprehensively calculate multiple positioning results, not only the reliability of the positioning result is improved, but also differentiated positioning strategies under different fault types are realized, so that the system can select the optimal positioning algorithm parameters according to the specific fault characteristics, significantly improving the positioning accuracy in complex fault scenarios.

[0016] Overall, the present invention forms a complete fault positioning chain from signal preprocessing, feature extraction to intelligent identification, and realizes the integrated intelligent monitoring of cable main core and sheath faults. Compared with the traditional separate detection system, it not only reduces the equipment complexity and maintenance cost, but also has the ability of adaptive learning and continuous optimization through the introduction of artificial intelligence algorithms, providing important technical support for the intelligent development of power system fault detection technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 A schematic flow chart of a cable fault location method based on neural network learning provided by an embodiment of the present invention; Figure 2 A schematic structural diagram of a cable fault location system based on neural network learning is provided in an embodiment of the present invention.

[0019] Reference numerals: 100, pre-processing module; 200, acquisition module; 300, extraction module; 400, recognition module; 500, positioning module. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the embodiments described are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0021] The embodiments of the present invention are described below with reference to the accompanying drawings.

[0022] like Figure 1 As shown, the present invention provides a cable fault location method based on neural network learning, comprising: S1: collecting a cable traveling wave signal and preprocessing the cable traveling wave signal to obtain a preprocessed traveling wave signal.

[0023] Wherein, step S1 further includes: S11: Collect cable traveling wave signals.

[0024] Furthermore, the cable traveling wave signal is collected through a high-frequency voltage transformer and a high-frequency current transformer. During the collection, sensors are installed at both ends of the cable (i.e., the substation end and the user end) to capture the traveling waves propagating in both directions. For long cables, monitoring points can be added at the middle joints to provide a reliable data basis for subsequent fault location.

[0025] S12: performing multi-band filtering on the cable traveling wave signal through a high-pass filter and a low-pass filter to obtain a filtered traveling wave signal.

[0026] Wherein, step S12 further includes: S121: Input the cable traveling wave signal into a high-pass filter to filter out baseband interference and obtain a high-frequency traveling wave component.

[0027] Furthermore, the fundamental frequency interference filtering in step S121 utilizes the frequency selection characteristics of the high-pass filter to eliminate the 50Hz power frequency interference and its low-frequency harmonic components that are prevalent in the power system. The high-pass filter adopts a steep cutoff frequency design and can effectively suppress the 50Hz fundamental frequency and its second harmonic 100Hz, while retaining the high-frequency transient characteristics in the fault traveling wave signal. When the cable traveling wave signal is input into the high-pass filter, the filter performs frequency domain processing according to its transfer function. For signal components below the cutoff frequency, the filter output amplitude is attenuated at a slope of 20dB / decade, while for signal components above the cutoff frequency, the original amplitude is basically maintained. After high-pass filtering, the amplitude of the 50Hz component is attenuated to less than 1 / 100 of the original value, and the high-frequency transient components generated by the fault are highlighted.

[0028] S122: Input the high-frequency traveling wave component into a low-pass filter for anti-aliasing filtering to obtain a band-limited filtered traveling wave signal.

[0029] Furthermore, in step S122, the present invention performs anti-aliasing filtering on the high-frequency traveling wave component output from step S121 to prevent spectral aliasing of the high-frequency components in the signal during subsequent ADC sampling. During data processing, when the high-frequency traveling wave component passes through the elliptical low-pass filter, the filter processes it according to its complex transfer function, producing deep attenuation in the stopband while controlling the amplitude-frequency characteristics in the passband. During the filtering process, each sampling point of the input signal is processed sequentially by four biquadratic sections, with the output of each section serving as the input to the next section, ultimately resulting in a band-limited filtered traveling wave signal.

[0030] S13: performing noise suppression on the filtered traveling wave signal by using an LMS adaptive filtering algorithm to obtain a preprocessed traveling wave signal.

[0031] Furthermore, in step S13, the present invention further eliminates noise from the band-limited filtered traveling wave signal using a least mean square adaptive filtering algorithm, aiming to minimize the mean square error between the desired signal and the filter output by continuously adjusting the filter weight coefficients. During the processing, the statistical characteristics of the signal are first used to estimate the autocorrelation function of the noise. Then, the optimal weight vector is calculated based on the Wiener filtering theory. Then, through iterative updates, the optimal solution is gradually approached. Finally, after iterations, the weight vector converges to a value near the optimal value. The obtained preprocessed traveling wave signal can also effectively suppress the noise components in the input signal while retaining the useful fault characteristic signal.

[0032] S2: Perform high-speed parallel acquisition on the pre-processed traveling wave signals to obtain multi-channel digital traveling wave data.

[0033] Wherein, step S2 further includes: S21: Distribute the pre-processed traveling wave signal to multiple LVDS differential output ADC chips for synchronous digital conversion to obtain a parallel digital sequence.

[0034] Furthermore, in step S21, the pre-processed traveling wave signal is first distributed to multiple LVDS differential output ADC chips. After passing through the signal distributor, the pre-processed traveling wave signal is simultaneously sent to multiple ADC chips working in parallel. Each ADC chip continuously samples the input signal. During the sampling process, the comparator circuit inside the ADC chip compares the continuous analog voltage value with the reference voltage, completes the analog-to-digital conversion through the successive approximation register architecture, and converts the voltage amplitude of each sampling point into a corresponding digital code value. The multiple ADC chips work simultaneously to form a parallel digital sequence. Each digital sequence finally obtained contains timestamp information and corresponding digitized amplitude data.

[0035] S22: Perform clock synchronization calibration on the parallel digital sequence to obtain the calibration traveling wave data.

[0036] Furthermore, the clock synchronization calibration process is intended to ensure multi-channel data. Since there are sampling clock deviations and signal propagation delay differences between different ADC chips, these errors need to be eliminated through a precise clock calibration algorithm. Therefore, in step S22, the present invention determines the time delay difference between each channel through cross-correlation calculation, and then uses a linear interpolation method to resample the time axis to align the data of all channels to a unified time base to eliminate time errors. The calibrated traveling wave data is aligned with nanosecond precision on the time axis.

[0037] S23: The calibration traveling wave data is stored in a high-speed cache, and valid segments of the cached data are extracted according to a preset trigger condition to obtain multi-channel digitized traveling wave data related to the fault event.

[0038] Furthermore, in step S23, after a circular buffer structure is used to calibrate the cache of the traveling wave data, a trigger condition detection algorithm is used to monitor abnormal features in the cached data in real time, including fault signs such as amplitude mutation, spectrum anomaly, and energy concentration. When it is detected that the amplitude of the traveling wave signal exceeds a preset threshold or the signal energy increases sharply in a short period of time, the trigger mechanism immediately starts the data extraction program. The specific trigger algorithm adopts a sliding window detection method to calculate the statistical characteristic parameters of the signal within a preset time window, including mean, variance, peak factor, etc. When the parameters deviate from the normal range, it is determined that a fault event has occurred, and all multi-channel data in a specific time period before and after the fault occurs are extracted. The final multi-channel digitized traveling wave data is the traveling wave data corresponding to the time period including the fault event.

[0039] S3: performing multi-resolution decomposition on the multi-channel digitized traveling wave data by means of a fast wavelet transform algorithm, and performing feature extraction on the wavelet coefficients obtained by the decomposition to obtain a fault feature vector.

[0040] Wherein, step S3 further includes: S31: Selecting a Daubechies wavelet basis, performing multi-resolution and multi-scale decomposition on the multi-channel digitized traveling wave data, and obtaining wavelet coefficients corresponding to multiple frequency bands.

[0041] Furthermore, the Daubechies wavelet basis function has a limited support length and good frequency selectivity, making it suitable for analyzing transient mutation characteristics in cable traveling wave signals. The Daubechies wavelet basis extracts the signal's approximate and detail components through the included low-pass and high-pass decomposition filters. Each layer of decomposition decomposes the signal into low-frequency approximate coefficients and high-frequency detail coefficients. Specifically, the multi-channel digitized traveling wave data obtained in the above steps are first decomposed into a five-layer decomposition using the Daubechies4 wavelet basis. The first layer of decomposition divides the original signal frequency band into an approximate component A1 of 0-32.5 MHz and a detail component D1 of 32.5-65 MHz. The second layer continues to decompose A1 to obtain A2 of 0-16.25 MHz and D2 of 16.25-32.5 MHz, and so on until the fifth layer, finally obtaining wavelet coefficients of 6 different frequency bands: A5 (0-2.03 MHz), D5 (2.03-4.06 MHz), D4 (4.06-8.125 MHz), D3 (8.125-16.25 MHz), D2 (16.25-32.5 MHz), and D1 (32.5-65 MHz). The wavelet coefficients of each frequency band reflect the time-frequency characteristics of the traveling wave signal within the corresponding frequency range.

[0042] S32: Perform adaptive threshold denoising on the wavelet coefficients of multiple frequency bands to obtain a denoised wavelet coefficient matrix.

[0043] Furthermore, the present invention adopts different threshold strategies for wavelet coefficients in different frequency bands. The specific threshold processing adopts a soft threshold function. When the absolute value of the wavelet coefficient is greater than the threshold, the coefficient value minus the threshold sign remains unchanged. When the absolute value of the wavelet coefficient is less than or equal to the threshold, the coefficient value is set to zero. The denoised wavelet coefficient matrix retains the main characteristic components of the signal while suppressing high-frequency noise interference, forming a denoised wavelet coefficient sequence corresponding to 6 frequency bands.

[0044] S33: Perform feature extraction on the denoised wavelet coefficient matrix to obtain a fault feature vector.

[0045] Wherein, step S33 further includes: S331: Calculate the time domain statistical characteristics of the denoised wavelet coefficient matrix to extract the first-order characteristics of the denoised wavelet coefficient matrix and obtain a time domain feature set; S332: Calculate the frequency domain energy distribution of the denoised wavelet coefficient matrix to extract the second-order characteristics of the denoised wavelet coefficient matrix and obtain a frequency domain feature set.

[0046] Furthermore, the first-order statistical characteristic parameters, including mean, standard deviation, skewness, kurtosis, energy, zero-crossing rate and other parameters, reflect the amplitude distribution characteristics and time-domain variation patterns of the traveling wave signal in each frequency band. For the denoised wavelet coefficients of the six frequency bands, the above six statistical characteristic parameters are calculated respectively to form a total of 36-dimensional time-domain feature sets; while the frequency domain energy distribution feature extraction is to obtain the second-order characteristic parameters by calculating the energy distribution and spectral characteristics of the wavelet coefficients of each frequency band. First, the relative energy of each frequency band is calculated, that is, the ratio of the wavelet coefficient energy of the frequency band to the total energy. Then, the energy ratio characteristics between frequency bands are calculated, including parameters such as the energy ratio of adjacent frequency bands, the high-low frequency energy ratio, and the energy ratio of specific frequency band combinations. These ratio characteristics reflect the spectral distribution characteristics of the fault traveling wave signal. Finally, for the energy distribution of the six frequency bands, 15 frequency domain characteristic parameters are calculated to form a 15-dimensional frequency domain feature set.

[0047] S333: Fusing the time domain feature set and the frequency domain feature set to obtain a fault feature vector.

[0048] In step S333, the present invention combines the 36-dimensional time domain feature set and the 15-dimensional frequency domain feature set, and first standardizes each type of feature. The standardized eigenvalues ​​have zero mean and unit variance, eliminating the dimensional differences between different features. Subsequently, the present invention uses the principal component analysis method to reduce the dimensionality of the 51-dimensional features, calculates the feature covariance matrix, solves the eigenvectors and eigenvalues, and finally selects the top k principal components with a cumulative contribution rate of 95% as the final fault feature vector.

[0049] S4: Classify and identify the fault feature vectors through a multi-layer perceptron neural network, and calculate the fault location based on the classification results to obtain a fault location result.

[0050] Wherein, step S4 further includes: S41: Input the fault feature vector into a multi-layer perceptron neural network to obtain a fault type probability distribution.

[0051] Wherein, step S41 further includes: S411: Perform forward propagation calculation on the fault feature vector through a multi-layer perceptron neural network to obtain a hidden layer activation output; S412: Perform Softmax normalization processing on the hidden layer activation output to obtain a fault type probability distribution.

[0052] Furthermore, the multilayer perceptron neural network adopted in the present invention is a feedforward network structure, including three main layers: input layer, hidden layer and output layer. The input layer receives a 23-dimensional fault feature vector, the hidden layer adopts a two-layer structure containing 64 and 32 neuron nodes respectively, and the output layer contains 5 neuron nodes corresponding to 5 fault types (single-phase grounding, two-phase short circuit, two-phase grounding, three-phase short circuit, and three-phase grounding).

[0053] During the forward propagation calculation process, the 23 eigenvalues ​​of the input layer are matrix multiplied with the weight matrix connected to the first hidden layer. The dimension of the weight matrix is ​​23×64, and the result of the matrix multiplication is a 64-dimensional vector. The bias vector is then added to obtain the linear combination value of the first hidden layer.

[0054] The activation function of the first hidden layer uses the ReLU function. The output for negative values ​​is 0, and the original value remains unchanged for positive values. After activation, the output of the first hidden layer is obtained. The obtained 64-dimensional vector is used as the input of the second hidden layer to continue the same calculation process. The weight matrix dimension of the second hidden layer is 64×32, and a 32-dimensional activation output is calculated.

[0055] The 32-dimensional activation output of the second hidden layer is connected to the output layer for the final classification calculation. The dimension of the output layer weight matrix is ​​32×5. The linear combination calculation obtains a 5-dimensional vector, and each element of the vector corresponds to the original score value of a fault type.

[0056] Softmax normalization converts the original score value into a probability distribution. The normalized output value range is between 0 and 1, and the sum of all output values ​​is 1. The Softmax calculation process first calculates the exponential value for each original score, then calculates the sum of all exponential values ​​as the normalization factor, and finally divides each exponential value by the normalization factor to obtain the corresponding probability value. The obtained probability value represents the confidence that the input feature vector belongs to the corresponding fault type.

[0057] S42: Determine the dominant fault type according to the fault type probability distribution and obtain a fault type identifier.

[0058] Furthermore, the fault type determination process in step S42 determines the dominant fault type by comparing the size relationship of the five probability values, selects the fault type with the largest probability value as the final classification result, and records the probability value as the classification confidence index. The fault type identification adopts a digital coding method. According to the aforementioned fault type, the single-phase grounding fault is coded as 1, the two-phase short circuit fault is coded as 2, the two-phase grounding fault is coded as 3, the three-phase short circuit fault is coded as 4, and the three-phase grounding fault is coded as 5. This coding value is used as the input parameter for subsequent positioning calculations. It should be noted that when the maximum probability value is lower than the preset confidence threshold, it is marked as an unknown fault type and requires manual intervention for further analysis.

[0059] S43: Calculating the arrival time difference of the two-end traveling waves based on the multi-channel digitized traveling wave data to perform time difference measurement and obtain the traveling wave propagation time difference.

[0060] Furthermore, the expression of the traveling wave propagation time difference is: ; in, is the traveling wave propagation time difference, To monitor the horizontal distance between the terminal and the fault point, is the propagation speed of the traveling wave in the cable medium, which is related to the cable material properties and frequency. is the cable length, is the traveling wave frequency, To monitor the vertical distance between the terminal and the fault point, is the collective structure correction factor, taking into account the conductor cross-sectional shape, is the frequency-dependent correction factor.

[0061] The two-terminal traveling wave arrival time difference calculation is based on the time information in the multi-channel digitized traveling wave data. First, the precise moment when the traveling wave arrives at each monitoring point is identified. The traveling wave arrival moment is determined by detecting the mutation point of the signal amplitude. The moment detection algorithm adopts a sliding window energy detection method. When the signal energy at a certain moment exceeds K times the average energy, it is determined as the traveling wave arrival moment, where K is a preset detection threshold. The time difference finally obtained reflects the difference in traveling wave propagation time caused by the difference in distance from the fault point to the two monitoring terminals.

[0062] S44: Using a double-end positioning method, distance calculation is performed according to the traveling wave propagation time difference to obtain the fault positioning result.

[0063] The expression of the fault location result in step S44 is: ; in, To calculate the fault location result, is the traveling wave propagation time difference, is the base time difference, is the collective structure correction factor, taking into account the cable wave impedance, is the frequency-dependent correction factor, is the cable length, To monitor the vertical distance between the terminal and the fault point, is the traveling wave frequency, It is the horizontal distance between the monitoring terminal and the fault point.

[0064] Furthermore, the above-mentioned traveling wave propagation time difference is obtained through the aforementioned time difference measurement. The reference time difference represents the time required for the traveling wave to propagate the entire length of the cable in an ideal uniform medium. The fault location result finally calculated is the distance value, that is, the distance between the fault point and the original end.

[0065] like Figure 2 As shown, the present invention also provides a cable fault location system based on neural network learning, comprising: Preprocessing module 100: used for collecting cable traveling wave signals and preprocessing the cable traveling wave signals to obtain preprocessed traveling wave signals; An acquisition module 200 is configured as a multi-channel LVDS differential output ADC chip, and is used to perform high-speed parallel acquisition of the pre-processed traveling wave signal to obtain multi-channel digitized traveling wave data; Extraction module 300: used to perform multi-resolution decomposition on the multi-channel digitized traveling wave data using a fast wavelet transform algorithm, and perform feature extraction on the wavelet coefficients obtained by the decomposition to obtain a fault feature vector; An identification module 400, configured as a multi-layer perceptron neural network, for classifying and identifying the fault feature vector to obtain a classification result; Positioning module 500: used to calculate the fault location based on the classification result to obtain a fault location result.

[0066] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0067] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0068] The present invention provides a cable fault location method and system based on neural network learning, which significantly improves the accuracy and reliability of cable fault location. Compared with traditional methods, the positioning error accuracy is improved, and the fault type misjudgment rate is effectively reduced. Multi-resolution wavelet decomposition and adaptive threshold denoising technology also improve the signal-to-noise ratio and significantly enhance the detection capability of weak fault signals. The dual-end positioning algorithm is combined with multiple correction factors, and can still maintain high positioning accuracy in complex cable laying environments and has strong adaptability. The overall solution shortens the fault detection response time to milliseconds, greatly improves the real-time performance of fault processing, and provides reliable protection for the safe and stable operation of the power system.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A cable fault location method based on neural network learning, characterized in that: include: S1: collecting a cable traveling wave signal and preprocessing the cable traveling wave signal to obtain a preprocessed traveling wave signal; S2: performing high-speed parallel acquisition on the pre-processed traveling wave signal to obtain multi-channel digital traveling wave data; S3: performing multi-resolution decomposition on the multi-channel digitized traveling wave data using a fast wavelet transform algorithm, and performing feature extraction on the wavelet coefficients obtained by the decomposition to obtain a fault feature vector; S4: Classify and identify the fault feature vector through a multi-layer perceptron neural network, and calculate the fault location based on the classification result to obtain a fault location result.

2. A cable fault location method based on neural network learning according to claim 1, characterized in that: Step S1 further comprises: S11: collect cable traveling wave signals; S12: performing multi-band filtering on the cable traveling wave signal through a high-pass filter and a low-pass filter to obtain a filtered traveling wave signal; S13: performing noise suppression on the filtered traveling wave signal by using an LMS adaptive filtering algorithm to obtain a preprocessed traveling wave signal.

3. A cable fault location method based on neural network learning according to claim 2, characterized in that: Step S12 further includes: S121: Input the cable traveling wave signal into a high-pass filter to filter out baseband interference and obtain a high-frequency traveling wave component; S122: Input the high-frequency traveling wave component into a low-pass filter for anti-aliasing filtering to obtain a band-limited filtered traveling wave signal.

4. A cable fault location method based on neural network learning according to claim 1, characterized in that: Step S2 further comprises: S21: Distributing the pre-processed traveling wave signal to multiple LVDS differential output ADC chips for synchronous digital conversion to obtain a parallel digital sequence; S22: Perform clock synchronization calibration on the parallel digital sequence to obtain the calibration traveling wave data; S23: The calibration traveling wave data is stored in a high-speed cache, and valid segments of the cached data are extracted according to a preset trigger condition to obtain multi-channel digitized traveling wave data related to the fault event.

5. The cable fault location method based on neural network learning according to claim 1, characterized in that: Step S3 further comprises: S31: Selecting a Daubechies wavelet basis, performing multi-resolution and multi-scale decomposition on the multi-channel digitized traveling wave data, and obtaining wavelet coefficients corresponding to multiple frequency bands; S32: performing adaptive threshold denoising on the wavelet coefficients of multiple frequency bands to obtain a denoised wavelet coefficient matrix; S33: Perform feature extraction on the denoised wavelet coefficient matrix to obtain a fault feature vector.

6. A cable fault location method based on neural network learning according to claim 5, characterized in that: Step S33 further includes: S331: Calculating the time-domain statistical features of the denoised wavelet coefficient matrix to extract the first-order features of the denoised wavelet coefficient matrix and obtain a time-domain feature set; S332: Calculating the frequency domain energy distribution of the denoised wavelet coefficient matrix to extract the second-order features of the denoised wavelet coefficient matrix and obtain a frequency domain feature set; S333: Fusing the time domain feature set and the frequency domain feature set to obtain a fault feature vector.

7. The cable fault location method based on neural network learning according to claim 1, characterized in that: Step S4 further comprises: S41: Inputting the fault feature vector into a multilayer perceptron neural network to obtain a fault type probability distribution; S42: Determine the dominant fault type according to the fault type probability distribution and obtain a fault type identifier; S43: Calculating the arrival time difference of the traveling waves at both ends based on the multi-channel digitized traveling wave data to perform time difference measurement and obtain the traveling wave propagation time difference; S44: Using a double-end positioning method, distance calculation is performed according to the traveling wave propagation time difference to obtain the fault positioning result.

8. The cable fault location method based on neural network learning according to claim 7, characterized in that: Step S41 further includes: S411: Perform forward propagation calculation on the fault feature vector through a multilayer perceptron neural network to obtain a hidden layer activation output; S412: Performing Softmax normalization processing on the hidden layer activation output to obtain a fault type probability distribution.

9. The cable fault location method based on neural network learning according to claim 7, characterized in that: The expression of the fault location result in step S44 is: ； in, The fault location result obtained by calculation is: is the traveling wave propagation time difference, is the base time difference, is the propagation speed of the traveling wave in the cable medium, is the collective structure correction coefficient, is the frequency-dependent correction factor, is the cable length, To monitor the vertical distance between the terminal and the fault point, is the traveling wave frequency, It is the horizontal distance between the monitoring terminal and the fault point.

10. A cable fault location system based on neural network learning, used to execute the cable fault location method based on neural network learning according to any one of claims 1 to 9, characterized in that: include: Preprocessing module: used for collecting cable traveling wave signals and preprocessing the cable traveling wave signals to obtain preprocessed traveling wave signals; An acquisition module, configured as a multi-channel LVDS differential output ADC chip, for performing high-speed parallel acquisition of the pre-processed traveling wave signal to obtain multi-channel digitized traveling wave data; Extraction module: used to perform multi-resolution decomposition on the multi-channel digitized traveling wave data through a fast wavelet transform algorithm, and perform feature extraction on the wavelet coefficients obtained by the decomposition to obtain a fault feature vector; an identification module, wherein the identification module is configured as a multi-layer perceptron neural network and is used to classify and identify the fault feature vector to obtain a classification result; Positioning module: used to calculate the fault location based on the classification result to obtain the fault location result.

Images