Distribution network cable fault early warning method and system based on multi-terminal synchronous traveling wave detection

Abstract

The invention discloses a distribution network cable fault early warning method and system based on multi-terminal synchronous traveling wave detection, and relates to the technical field of cable fault detection. The method comprises the following steps: deploying at least three synchronous detection ends at each branch end point of a cable, and synchronously triggering a sampling rate; establishing a master-slave synchronous network, and performing frequency spectrum correction on the acquired data; constructing a real-time correction model based on the data of the distributed temperature sensor; extracting early fault features, and constructing a three-dimensional feature tensor; carrying out preliminary diagnosis on edge nodes and carrying out feature classification by adopting a transfer learning optimization network; and in combination with the cable load rate and the environment temperature and humidity, an early warning threshold self-adaptive model based on fuzzy PID (Proportion Integration Differentiation) is established. Through multi-terminal cooperative perception and intelligent analysis, the scheme converts a fault processing mode from passive first-aid repair to active defense, improves the manual inspection efficiency of the distribution network cable, and finds early fault hidden dangers in advance.

Description

Technical Field

[0001] The present invention belongs to the technical field of cable fault detection, and particularly relates to a distribution network cable fault early warning method and system based on multi-terminal synchronous traveling wave detection. Background Art

[0002] With the continuous expansion of the urban power grid scale, high-voltage power cables have become the core carriers of the urban power transmission and distribution system. However, during the long-term operation of cables, potential faults such as insulation degradation and partial discharge caused by factors such as load fluctuations, environmental temperature and humidity changes, and mechanical stress are becoming increasingly prominent, seriously threatening power supply reliability. In the existing technology, cable condition monitoring and fault early warning face the following key bottlenecks; Threshold rigidity problem: Traditional monitoring systems mostly adopt fixed threshold criteria (such as partial discharge amount > 50 pC, temperature rise rate > 3 °C / min), but the actual operating state of cables is strongly coupled by dynamic loads and environmental factors. For example, in a high-humidity environment in summer, the insulation performance of cables will decrease as the humidity increases, and at this time, the fixed threshold may lead to missed alarms; while in a low-temperature and high-load condition in winter, false alarms are easily caused due to overly conservative thresholds.

[0003] Limitations in feature extraction: Existing fault feature extraction methods (such as wavelet transform and spectrum analysis) are easily affected by power frequency harmonics and electromagnetic interference in a strong noise environment. Typical cases show that when the cable load rate > 80%, the capture rate of the traditional wavelet packet decomposition algorithm for early partial discharge features drops to 61.3%, and there is also a phenomenon of feature aliasing caused by spectrum leakage.

[0004] Insufficient accuracy in synchronous acquisition: Multi-terminal traveling wave positioning technology relies on high-precision time synchronization, but existing systems mostly adopt a single GPS clock synchronization (accuracy ±200 ns), which is difficult to meet the sub-microsecond-level synchronization requirements in a complex electromagnetic environment. The analysis of a 500 kV cable fault case shows that the positioning error caused by synchronous deviation reaches ±300 meters, seriously delaying the fault repair time.

[0005] Environmental adaptability defects: In extreme temperature and humidity conditions (such as -40 °C to +70 °C, humidity > 95%RH), sensors of traditional monitoring terminals are prone to problems such as baseline drift and sampling inaccuracy. Monitoring data of a cable network in a plateau area shows that in a low-temperature environment in winter, the signal-to-noise ratio (SNR) of vibration sensors drops by about 15 dB, resulting in an increase in the missed detection rate of mechanical faults to 28%.

[0006] In view of the above problems, the present invention constructs a multi-dimensional collaborative monitoring system for cable conditions by deeply integrating dynamic parameter perception, adaptive feature extraction, and high-precision synchronous acquisition technologies, aiming to break through the bottlenecks of existing technologies and achieve precise early warning of cable hidden dangers. Summary of the Invention

[0007] The object of the present invention is to provide a distribution network cable fault warning method and system based on multi-terminal synchronous traveling wave detection. Through multi-terminal collaborative perception and intelligent analysis, this solution changes the fault handling mode from "passive repair" to "active defense", and solves the problems of low efficiency of manual inspection of existing distribution network cables, high concealment of early fault hazards, and difficulty in discovery.

[0008] To solve the above technical problems, the present invention is realized through the following technical solutions: The present invention is a distribution network cable fault warning method based on multi-terminal synchronous traveling wave detection, including the following steps: Step S1: Deploy at least three synchronous detection terminals at the endpoints of each branch of the cable, and trigger the sampling rate synchronously; Step S2: Establish a master-slave synchronous network, use FPGA to generate a sampling clock to realize multi-terminal synchronous startup, and perform spectrum correction on the collected data; Step S3: Establish a traveling wave propagation matrix model of the cable network, and calculate the direction of wave arrival through the cross-correlation time difference method; Step S4: Based on the data of the distributed temperature sensor, construct a real-time correction model; Step S5: Extract early fault features and construct a three-dimensional feature tensor; Step S6: The edge node performs preliminary diagnosis and uses transfer learning to optimize the network for feature classification; Step S7: Combine the cable load rate with the ambient temperature and humidity to establish an adaptive warning threshold model based on fuzzy PID.

[0009] As a preferred technical solution, in the step S1, the synchronous detection terminals are deployed at the beginning and end and the intermediate joints of the main cable and the T-junction branch to form a multi-terminal monitoring array; the synchronous detection terminals include high-frequency current transformers and piezoelectric vibration sensors; the high-frequency current transformers are used to collect the core wire current and the electromagnetic field of the shielding layer. By detecting the discharge pulse current signals flowing through the cable grounding wire, neutral point wiring, and the cable body, the high-frequency current transformers can capture the high-frequency current signals generated during partial discharge inside the cable. These signals propagate along the outer shielding layer or grounding wire of the cable. As a "signal catcher", the high-frequency current transformers can accurately capture these weak signals and convert them into measurable electrical signals, providing a data basis for subsequent analysis and processing; the piezoelectric vibration sensors are used to collect mechanical vibration signals, mainly including vibration acceleration, vibration frequency, and vibration amplitude, which will generate charges. By measuring these charges or the voltage signals converted from them, the magnitude of the vibration acceleration can be determined. In addition, the sensors can also measure parameters such as the frequency and amplitude of the vibration.

[0010] As a preferred technical solution, in step S2, a master-slave synchronization network is established based on Beidou / GPS; wherein, the master clock node periodically sends 1PPS second pulses, and the slave nodes transmit PTPv2 protocol messages through the cable sheath to achieve a time deviation of less than 50ns for all nodes; the specific process of spectrum correction is as follows: Step S21: Use a parameterized Vincent window to truncate the signal, and the specific formula is as follows: ; In the formula, is a dynamic adjustment factor, and its value range is [0.08, 0.16], which is adaptively adjusted by real-time monitoring of the background noise spectral density; is the weighted coefficient value of the window function at the discrete time point , is the discrete time index, is the total length of the window function, is the core of main lobe shaping, is the side lobe suppression adjustment; Step S22: Perform a 4096-point FFT on the windowed new signal to obtain the initial spectrum , and calculate the maximum amplitude point within the main lobe range; Step S23: Construct a cubic spline interpolation function within the interval to solve the true frequency deviation ; Step S24: Establish an amplitude-phase correction model , and perform iterative improvement of the frequency resolution. In the formula, is the corrected signal amplitude, is the original measured amplitude, is the frequency deviation factor, is the spectral leakage compensation term; Step S25: Real-time monitor the power frequency offset, establish a dynamic compensation matrix, and eliminate the influence of frequency fluctuations; Step S26: Construct a multi-filter architecture for anti-aliasing processing; the pre-stage of anti-aliasing processing uses an analog anti-aliasing filter (cut-off frequency 500kHz, transition band attenuation 120dB / oct), and the post-stage digital filter uses an elliptic filter (passband ripple <0.01dB, stopband attenuation >90dB); perform cubic polynomial frequency domain interpolation after spectrum correction to compensate for the amplitude attenuation caused by the window function, and the specific formula is as follows: ; In the formula, is the final amplitude after frequency domain interpolation compensation, is the current analyzed frequency point, is the actual frequency of the signal or the center frequency of the spectrum main lobe, is the sampling frequency.

[0011] As a preferred technical solution, in step S3, the specific process of establishing the cable network traveling wave propagation matrix model is as follows: Step S31: Construct a weighted graph of the distribution network according to the distribution network cable system ; In the formula, the vertex set represents physical nodes, and the edge set represents the parameters of the cable segment. The weight matrix stores the propagation delay and attenuation coefficient between each node; Step S32: Establish a time-varying propagation matrix , and obtain the transfer function during the traveling wave transmission process: ; In the formula, is the multipath reflection coefficient, is the path delay, is the cable frequency-variable loss transfer function; Step S33: Establish a joint differential equation for traveling wave propagation; Step S34: Perform continuous wavelet transform on the two signals and calculate the cross-correlation: ; In the formula, is the wavelet coefficient matrix, is the weight function, focused on the main frequency band, is the corresponding traveling wave characteristic scale; Step S35: Construct an interpolation function within ±3 sampling points of the cross-correlation peak, and the specific formula is: ; In the formula, is the cubic spline basis function, is the true time difference estimate value, is to find the value that makes the objective function reach the maximum, is the weighted sum of discrete sampling points, is the signal at the time delay of the cross-correlation function value, is the cubic spline basis function; Step S36: Construct a cost function to distinguish the direct wave and the reflected wave, and solve the direction of arrival; Through cubic spline interpolation, the discrete cross-correlation function is continuously reconstructed, breaking through the traditional sampling interval limit, and finally solving the exact time delay that makes the reconstructed cross-correlation function reach the peak .

[0012] As a preferred technical solution, in step S36, the direction-of-arrival calculation process is as follows: Step S361: Calculate the time difference between every two of the N terminals , forming a time-difference matrix ; Step S362: Use the propagation matrix to perform path inversion: ; In the formula, is the parameter vector or matrix to be optimized, is the actually observed time-difference matrix, is the propagation model matrix, is the square of the Frobenius norm, is the regularization coefficient, is the L1 norm; The role of direction-of-arrival calculation is to find the one that can not only make the time-difference model have the highest matching degree ( is the smallest), but also ensure the sparsity of the solution ( is the smallest) of . In cable fault detection, its solution corresponds to the most likely fault position and intensity distribution; Step S363: Construct the array manifold matrix , and solve the spatial spectrum peak value. The calculation formula is: ; In the formula, is the optimal direction of arrival, that is, the incident angle of the signal source. By maximizing the spatial spectrum function, the true direction of the signal source is determined, is the value for finding the maximum of the expression. In the angle search range, find the that makes the denominator the smallest (because the smaller the denominator, the larger the overall fraction), is the conjugate transpose of the steering vector, is the noise subspace matrix, which is obtained by performing eigenvalue decomposition on the covariance matrix; is the noise subspace projection matrix, which represents the projection of the signal on the noise subspace. When is orthogonal to the noise subspace, the projection energy of this matrix is the smallest, corresponding to the smallest denominator value; The specific role is to find the signal source direction by searching for the with the smallest denominator, When is orthogonal to the noise subspace When they are orthogonal, the denominator approaches zero, the fraction is the largest, corresponding to the true direction of the signal source; by utilizing the orthogonality between the noise subspace and the signal subspace, noise interference is suppressed and the direction estimation accuracy is improved.

[0013] As a preferred technical solution, in step S4, the material characteristics are analyzed as follows: Establish a regression equation for the dielectric constant of the cable insulating material and temperature: ; In the formula, the coefficient is measured through a thermal aging test, and T is the temperature; Derive the conductor expansion coefficient model: ; In the formula, is the expansion coefficient of the conductor, is the length of the object at temperature T, is the initial length at the reference temperature When fusing distributed temperature sensing and the temperature characteristics of cable materials, the error in wave velocity calculation is reduced, and the traveling wave positioning accuracy under complex working conditions is effectively improved.

[0014] As a preferred technical solution, in step S5, the specific process of extracting early fault characteristics and constructing a three-dimensional feature tensor is as follows: Step S51: Dynamically select the optimal wavelet basis function according to the cable fault characteristic frequency band, and construct a parameterized Morlet wavelet family: ; where is dynamically adjusted according to the instantaneous frequency of the signal; Step S52: Introduce a local-global weight factor and calculate the dynamic energy entropy ratio. The calculation formula is as follows: ; In the formula, is the wavelet energy entropy ratio of the kth node, is the wavelet transform coefficient, indicating the time-frequency characteristics of the signal at scale and translation parameter ; Step S53: Perform 6-layer wavelet packet decomposition to cover the 0-30 MHz frequency band. Calculate the normalized WEE value for each node to form a 64-dimensional feature vector, and retain the first 8 most discriminative features through KL divergence; Step S54: Calculate the time synchronization difference degree based on the time difference of arrival of traveling waves at multiple terminals: ; In the formula, is the arrival time of the wavefront marked by each terminal, and the time resolution is improved through the cross-correlation algorithm; Step S55: Integrate electromagnetic / vibration bimodal data to construct a hybrid spectrum matrix; where the row vector is the 1MHz spectrum data collected by each terminal (resolution 0.5Hz from 0 - 500kHz); the column vector is the time series from 10 seconds before the fault to 200ms after the fault. Step S56: Perform truncated SVD decomposition , retain the first 3 singular values , and the formula for the three-dimensional feature tensor is ; are the left singular vector matrix, singular value matrix, and right singular vector matrix respectively; the three-dimensional feature tensors are the time synchronization difference degree, modal energy distribution entropy, and spectral singular value decomposition coefficient respectively.

[0015] As a preferred technical solution, in step S6, the specific process of the edge node preliminary diagnosis is as follows: Step S61: Dynamically segment the real-time traveling wave signal using a sliding window mechanism, and the window length is adaptively adjusted according to the cable length; Step S62: Extract the third-order feature vectors of the segmented waveform: time-domain features, frequency-domain features, and morphological features; where the time-domain features include the crest factor and waveform factor; the frequency-domain features include the proportion of the main frequency band energy; the morphological features include the rising edge slope; Step S63: Introduce a dynamic time adjustment optimization algorithm to solve the problem of waveform time axis offset, and the specific formula is as follows: ; In the formula, the distance metric adopts a weighted combination of Manhattan distance and cosine similarity; an adaptive noise suppression layer is added, and Gaussian filtering is automatically activated when the signal-to-noise ratio (SNR) < 15dB to improve the comparison stability; Step S64: Establish a fault waveform feature library and perform a three-level similarity determination. The fault waveform feature library contains the reference waveforms of 16 typical fault modes (grounding / short circuit / open circuit, etc.); the three-level similarity determination includes a rough screening stage: calculate the similarity of the waveform envelope area (threshold > 85%); a fine screening stage: calculate the Hausdorff distance (threshold < 0.15); a verification stage: verify the polarity consistency of the waveforms of adjacent terminals.

[0016] As a preferred technical solution, in step S6, the network structure includes: constructing a 12-layer ResNet-1D structure, including: Input layer: Accept a 128-dimensional feature vector (64 dimensions in the time domain + 64 dimensions in the frequency domain); Residual block: Use dilated convolution (dilation = 2) to expand the receptive field; Attention mechanism: Embed the SE module to dynamically adjust the channel weights; When optimizing the network through transfer learning, perform spatio-temporal alignment on the multi-terminal data received by the cloud, execute improved SVD noise reduction processing, and retain the first three principal components; perform multi-scale feature fusion, specifically including: Low-level features: the first-order difference signal of the original waveform; Mid-level features: the node energy of wavelet packet decomposition; High-level features: the singular value coefficients of the time-frequency joint distribution; Adopt the triplet loss function , where is the boundary parameter, are the anchor point, positive sample, and negative sample respectively; when dynamically adjusting the class weights, automatically calculate the class weights based on the fault occurrence frequency (the weights of rare faults are increased by 3 to 5 times), and introduce a fuzzy membership function to process the boundary fuzzy samples.

[0017] As a preferred technical solution, in step S7, the specific steps for establishing an early warning threshold adaptive model based on fuzzy PID are as follows: Step S71: Measure the current of the cable core wire in real time through a Rogowski coil and calculate the load rate; ; where is the load rate of the cable at time t, is the effective value of the current at time t, is the rated current of the cable; Step S72: Deploy distributed temperature and humidity sensors, install 1 set every 200-meter cable section, and establish an environmental factor matrix , where respectively represent the outer surface temperature of the cable, the humidity of the environment where the cable is located, and the operating temperature of the cable core conductor. The conductor temperature is measured non-contact by an infrared thermal imager; Step S73: Define three fuzzy input quantities, including load rate deviation, temperature rise rate, and four-degree influence factor, establish a fuzzy rule base, with a total of 81 control rules, and the rule weights are dynamically adjusted according to the aging characteristics of the cable insulation material; Step S74: Self-tuning of PID parameters, output quantity defuzzification formula: ; where is the activation degree, is the output value corresponding to the rule, and the proportional-integral-differential coefficients are updated in real time to ensure that the adjustment period < 500 ms; Step S75: Establish an initial threshold matrix according to the cable model ; where are the partial discharge threshold, harmonic distortion rate, and traveling wave amplitude threshold respectively, and construct an adaptive adjustment equation, the specific formula is as follows: ; where is the reference threshold, is the proportionality coefficient, is the load rate deviation, is the integral coefficient, is the integral term of the temperature rise rate, is the differential coefficient, is the transformation rate of the temperature influence factor, restricting the correction amplitude within the range of ±40% to prevent overshoot; Step S76: Introduce the Sigmoid function for non - linear limiting to ensure that the threshold is always within the safe range and complete the dynamic calculation of the early warning threshold.

[0018] The present invention is a distribution network cable fault early warning system based on multi - terminal synchronous traveling wave detection, including a distributed traveling wave sensing array, a quantum time synchronization network, an intelligent analysis center, and an anti - interference transmission system; The distributed traveling wave sensing array includes a multi - modal sensor cluster and an edge computing node; the multi - modal sensor cluster deploys an electromagnetic - ultrasonic composite sensor at the cable joint, integrating a high - frequency current transformer and a piezoelectric vibration sensor for synchronous acquisition of multi - physical quantities of traveling waves; the edge computing node has an FPGA pre - processing chip built into each terminal for wavelet packet threshold denoising and traveling wave arrival time marking; The quantum time synchronization network includes a Beidou / GPS - rubidium atomic clock dual - mode timing module and a PTPv2 protocol optimization channel; the Beidou / GPS - rubidium atomic clock dual - mode timing module is used for the time synchronization error of each terminal; the PTPv2 protocol optimization channel is used to establish a hybrid communication link based on the cable metal sheath to ensure that the time synchronization signal can still be transmitted during communication interruption; The intelligent analysis center includes a feature fusion engine, a deep prediction model, and a dynamic risk assessment module; the feature fusion engine uses a multi - source data fusion algorithm with an attention mechanism to establish a time - domain - frequency - domain - energy - domain joint feature matrix; the deep prediction model is a spatio - temporal sequence prediction network based on TCN - GRU, embedding prior knowledge of cable topology for early fault feature extraction; the dynamic risk assessment module is used to construct a five - level early warning index system by combining the D - S fusion algorithm with improved evidence theory; The anti - interference transmission system is a dual - channel redundancy transmission of power line carrier and NB - IoT.

[0019] The present invention has the following beneficial effects: (1) By deploying at least three synchronous detection terminals at the endpoints of each branch of the cable, establishing a master - slave synchronous network, correcting the spectrum of the collected data to establish a traveling wave propagation matrix model of the cable network, and establishing an adaptive early warning threshold model based on fuzzy PID, the present invention realizes early warning of faults, effectively identifies early weak faults, and achieves a technological leap from passive positioning to active early warning through the innovative combination of multi - physical quantity fusion and deep learning technology.

[0020] (2) By integrating the cable network topology, material parameters, and environmental temperature and humidity into the time-varying propagation matrix, the present invention reduces the model error. Meanwhile, considering the propagation differences between electromagnetic traveling waves and mechanical stress waves, it realizes the calculation of the composite wave arrival direction. By introducing qubit coding into the particle swarm optimization algorithm, the DOA estimation resolution reaches 0.5°. Through the inversion of the propagation matrix and the constraint of the cost function, it effectively suppresses the false alarms caused by reflected waves and reduces the false alarm rate.

[0021] (3) Through the fuzzy PID threshold adjustment model, the present invention reduces the false alarm rate and missed alarm rate under complex working conditions such as ±40% load fluctuation and sudden changes in temperature and humidity (ΔT>10℃ / h, ΔH>30%RH / h). Based on the local-global weight factor and adaptive threshold noise reduction technology, it improves the separation degree of effective signals, which is improved compared with the traditional wavelet packet method.

[0022] (4) Through the three-dimensional feature tensor of time synchronization difference degree, modal energy distribution entropy, and spectral singular value, the present invention can simultaneously characterize the coupling effects of multiple physical fields such as electricity, mechanics, and thermodynamics, and improve the accuracy of fault type identification.

[0023] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0025] Figure 1 It is a flowchart of a distribution network cable fault warning method based on multi-terminal synchronous traveling wave detection of the present invention; Figure 2 It is a schematic structural diagram of a distribution network cable fault warning system based on multi-terminal synchronous traveling wave detection of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0027] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0028] To make the objectives, technical solutions, and advantages of this application clearer, the following will further describe the embodiments of this application in detail with reference to the appended Figure 2 drawings.

[0029] Before introducing the embodiments of this application, relevant explanations will first be made on the application of traveling waves on distribution network cables.

[0030] The traveling wave fault location technology is based on the traveling wave theory. When a fault occurs on a transmission line, sudden changes in voltage and current will occur at the fault point, forming traveling waves. The traveling waves propagate along the transmission line towards both ends, and their propagation characteristics are related to the characteristic impedance, wave impedance, and distributed parameters of the line. During the propagation process of traveling waves, reflections and transmissions will occur when encountering points of discontinuous line impedance (such as the line end, branch point, or fault point).

[0031] Traveling waves propagate along the transmission line at nearly the speed of light. Their propagation speed is related to the electrical parameters of the line and is usually between 290,000 and 310,000 kilometers per second. When the traveling waves reach the measurement points at both ends of the line, reflected waves and transmitted waves will be generated. By accurately measuring the time difference of the traveling waves arriving, combined with the wave speed of the line, the distance from the fault point to the measurement point can be calculated.

[0032] To make the objectives, technical solutions, and advantages of this application clearer and more understandable, the following will further describe this application in detail with reference to the appended Figure 1-2 drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0033] Embodiment 1. Please refer to Figure 1 As shown in the figure, the present invention is a method for early warning of distribution network cable faults based on multi-terminal synchronous traveling wave detection, including the following steps: Step S1: Deploy at least three synchronous detection terminals at the endpoints of each branch of the cable, and trigger the sampling rate synchronously; Step S2: Establish a master-slave synchronous network, use FPGA to generate a sampling clock to achieve multi-terminal synchronous startup, and perform spectral correction on the collected data; Step S3: Establish a traveling wave propagation matrix model for the cable network, and calculate the direction of wave arrival through the cross-correlation time difference method; Step S4: Based on the distributed temperature sensor data, construct a real-time correction model; Step S5: Extract early fault characteristics and construct a three-dimensional feature tensor; Step S6: The edge node performs preliminary diagnosis and uses transfer learning to optimize the network for feature classification; Step S7: Combine the cable load rate and the ambient temperature and humidity to establish an adaptive warning threshold model based on fuzzy PID.

[0034] In step S1, synchronous detection terminals are deployed at the head and tail and intermediate joints of the cable main line and the T-branch to form a multi-terminal monitoring array; the synchronous detection terminals include high-frequency current transformers and piezoelectric vibration sensors; the high-frequency current transformers are used to collect the core wire current and the electromagnetic field of the shielding layer. By detecting the discharge pulse current signals flowing through the cable grounding wire, neutral point wiring, and the cable body, the high-frequency current transformers can capture the high-frequency current signals generated during partial discharge inside the cable. These signals propagate along the outer shielding layer or grounding wire of the cable. As a "signal catcher", the high-frequency current transformers can accurately capture these weak signals and convert them into measurable electrical signals, providing a data basis for subsequent analysis and processing; the piezoelectric vibration sensors are used to collect mechanical vibration signals, mainly including vibration acceleration, vibration frequency, and vibration amplitude, which generate charges. By measuring these charges or the voltage signals converted from them, the magnitude of the vibration acceleration can be determined. In addition, the sensor can also measure parameters such as the frequency and amplitude of the vibration.

[0035] In step S2, a master-slave synchronous network is established based on Beidou / GPS; among them, the master clock node periodically sends 1PPS second pulses, and the slave nodes transmit PTPv2 protocol messages through the cable sheath to achieve a time deviation of less than 50ns for all nodes; the specific process of spectrum correction is as follows: Step S21: Use a parameterized Vincent window for signal truncation, and the specific formula is as follows: ; In the formula, is a dynamic adjustment factor, and its value range is [0.08, 0.16], which is adaptively adjusted by real-time monitoring of the background noise spectral density; is the weighted coefficient value of the window function at the discrete time point , is the discrete time index, is the total length of the window function, is the core of main lobe shaping, is the side lobe suppression adjustment; Step S22: Perform a 4096-point FFT on the windowed new signal to obtain the initial spectrum , and calculate the maximum amplitude point within the main lobe range; Step S23: Construct a cubic spline interpolation function within the interval to solve the true frequency deviation ; Step S24: Establish an amplitude-phase correction model , and perform iterative improvement of the frequency resolution. In the formula, is the corrected signal amplitude, is the original measured amplitude, is the frequency deviation factor, is the spectral leakage compensation term; Step S25: Monitor the power frequency offset in real time, establish a dynamic compensation matrix, and eliminate the influence of frequency fluctuations; Step S26: Construct a multi-stage filtering architecture for anti-aliasing processing; the pre-stage of the anti-aliasing processing uses an analog anti-aliasing filter (cut-off frequency 500 kHz, transition band attenuation 120 dB / oct), and the post-stage digital filter uses an elliptic filter (passband ripple <0.01 dB, stopband attenuation >90 dB); perform cubic polynomial frequency domain interpolation after spectral correction to compensate for the amplitude attenuation caused by the window function. The specific formula is as follows: ; In the formula, is the final amplitude after frequency domain interpolation compensation, is the current analyzed frequency point, is the actual frequency of the signal or the center frequency of the spectral main lobe, is the sampling frequency.

[0036] In step S3, the specific process of establishing the traveling wave propagation matrix model of the cable network is as follows: Step S31: Construct a weighted graph of the distribution network according to the cable system of the distribution network ; In the formula, the vertex set represents the physical nodes, and the edge set represents the parameters of the cable section, and the weight matrix stores the propagation delay and attenuation coefficient between each node; Step S32: Establish a time-varying propagation matrix , and obtain the transfer function during the traveling wave transmission process: ; In the formula, is the multipath reflection coefficient, is the path delay, is the cable frequency-variable loss transfer function; Step S33: Establish the joint differential equation of traveling wave propagation; Step S34: Perform continuous wavelet transform on the two signals and calculate the cross-correlation: ; In the formula, is the wavelet coefficient matrix, is the weight function, focused on the main frequency band, is the corresponding traveling wave characteristic scale; Step S35: Construct an interpolation function within ±3 sampling points of the cross-correlation peak. The specific formula is: ; In the formula, is the cubic spline basis function, is the true time difference estimate value, is to find the value that makes the objective function reach the maximum value, is to perform weighted summation on discrete sampling points, is the signal at the time delay the cross-correlation function value at, is the cubic spline basis function; Step S36: Construct a cost function to distinguish the direct wave from the reflected wave and solve the direction of arrival; Through cubic spline interpolation, the discrete cross-correlation function is continuously reconstructed, breaking through the traditional sampling interval limit, and finally solving the exact time delay that makes the reconstructed cross-correlation function reach the peak .

[0037] In step S36, the direction-of-arrival calculation process is as follows: Step S361: Calculate the time difference for each pair of N terminals, to form a time difference matrix ; Step S362: Use the propagation matrix for path inversion: In the formula, is the parameter vector or matrix to be optimized, is the actually observed time difference matrix, is the propagation model matrix, is the square of the Frobenius norm, is the regularization coefficient, is the L1 norm; The role of the direction-of-arrival calculation is to find the that can not only make the time difference model have the highest matching degree ( is the smallest), but also ensure the sparsity of the solution ( is the smallest). In cable fault detection, its solution corresponds to the most likely fault location and intensity distribution; Step S363: Construct an array manifold matrix , solve the spatial spectrum peak value, and the calculation formula is: ; In the formula, is the optimal direction of arrival, that is, the incident angle of the signal source, is to find the value of the maximum value of the expression, is the conjugate transpose of the steering vector, is the noise subspace matrix, is the noise subspace projection matrix, representing the projection of the signal onto the noise subspace; its specific function is to find the direction of the signal source by searching for the smallest denominator in .

[0038] In step S4, analyze the material characteristics: Establish the regression equation of the dielectric constant of the cable insulation material and temperature: ; In the formula, the coefficient is measured through a thermal aging test, and T is the temperature; Derive the conductor expansion coefficient model: ; In the formula, is the expansion coefficient of the conductor, is the length of the object at temperature T, is the initial length at the reference temperature . By fusing distributed temperature sensing and the temperature characteristics of cable materials, the error in wave velocity calculation is reduced, and the traveling wave positioning accuracy under complex working conditions is effectively improved.

[0039] In step S5, extract the early fault characteristics, and the specific process of constructing the three-dimensional feature tensor is as follows: Step S51: Dynamically select the optimal wavelet basis function according to the cable fault characteristic frequency band, and construct a parameterized Morlet wavelet family: ; where is dynamically adjusted according to the instantaneous frequency of the signal; Step S52: Introduce a local-global weight factor and calculate the dynamic energy entropy ratio. The calculation formula is as follows: ; In the formula, is the wavelet energy entropy ratio of the kth node, is the wavelet transform coefficient, representing the time-frequency characteristics of the signal at scale and translation parameter ; Step S53: Perform 6-layer wavelet packet decomposition, covering the 0-30 MHz frequency band. Calculate the normalized WEE value for each node to form a 64-dimensional feature vector, and retain the first 8 most discriminative features through KL divergence; Step S54: Calculate the time synchronization difference degree based on the time difference of arrival of traveling waves at multiple terminals: ; In the formula, The arrival time of the wavefront marked for each terminal is improved in time resolution through the cross-correlation algorithm; Step S55: Fuse the electromagnetic / vibration dual-modal data to construct a hybrid spectral matrix; where the row vector is the 1MHz spectral data (resolution 0.5Hz from 0 - 500kHz) collected by each terminal; the column vector is the time series from 10 seconds before the fault to 200ms after the fault; Step S56: Perform truncated SVD decomposition , retain the first 3 singular values , and the formula for forming the three-dimensional feature tensor is ; are the left singular vector matrix, the singular value matrix, and the right singular vector matrix respectively; the three-dimensional feature tensor is the time synchronization difference degree, the modal energy distribution entropy, and the spectral singular value decomposition coefficient respectively.

[0040] In step S6, the specific process of the preliminary diagnosis by the edge node is as follows: Step S61: Use the sliding window mechanism to dynamically segment the real-time traveling wave signal, and the window length is adaptively adjusted according to the cable length; Step S62: Extract the third-order feature vectors of the segmented waveform: time domain features, frequency domain features, and morphological features; among them, the time domain features include the crest factor and the waveform factor; the frequency domain features include the proportion of the main frequency band energy; the morphological features include the rising edge slope; Step S63: Introduce the dynamic time adjustment optimization algorithm to solve the problem of waveform time axis offset, and the specific formula is as follows: ; In the formula, the distance metric adopts the weighted combination of Manhattan distance and cosine similarity; add an adaptive noise suppression layer, and when the signal-to-noise ratio (SNR) < 15dB, Gaussian filtering is automatically activated to improve the comparison stability; Step S64: Establish a fault waveform feature library and perform a three-level similarity determination. The fault waveform feature library contains the reference waveforms of 16 types of typical fault modes (grounding / short circuit / open circuit, etc.); the three-level similarity determination includes the rough screening stage: calculate the similarity of the waveform envelope area (threshold > 85%); the fine screening stage: calculate the Hausdorff distance (threshold < 0.15); the verification stage: verify the polarity consistency of the waveforms of adjacent terminals.

[0041] In step S6, the network structure includes: construct a 12-layer ResNet-1D structure, including: Input layer: Accept a 128-dimensional feature vector (64 dimensions in the time domain + 64 dimensions in the frequency domain); Residual block: Use dilated convolution (dilation = 2) to expand the receptive field; Attention mechanism: Embed the SE module to dynamically adjust the channel weights; When performing transfer learning to optimize the network, spatio-temporal alignment is carried out on the multi-terminal data received by the cloud, improved SVD noise reduction processing is performed, and the first 3 principal components are retained; multi-scale feature fusion is carried out, specifically including: Low-level features: the first-order difference signal of the original waveform; Intermediate features: the node energy of wavelet packet decomposition; High-level features: the singular value coefficients of the time-frequency joint distribution; The triplet loss function is adopted , where is the boundary parameter, are the anchor point, positive sample, and negative sample respectively; when dynamically adjusting the class weights, the class weights are automatically calculated based on the fault occurrence frequency (the weights of rare faults are increased by 3 to 5 times), and a fuzzy membership function is introduced to process the boundary fuzzy samples.

[0042] In step S7, the specific steps for establishing an adaptive early warning threshold model based on fuzzy PID are as follows: Step S71: Measure the current of the cable core wire in real time through a Rogowski coil and calculate the load rate; ; where is the load rate of the cable at time t, is the effective value of the current at time t, is the rated current of the cable; Step S72: Deploy distributed temperature and humidity sensors, install 1 set every 200-meter cable section, and establish an environmental factor matrix , where respectively represent the outer surface temperature of the cable, the humidity of the environment where the cable is located, and the operating temperature of the cable core conductor. The conductor temperature is measured non-contact through an infrared thermal imager; Step S73: Define three fuzzy input quantities, including load rate deviation, temperature rise rate, and four-degree influence factor, establish a fuzzy rule base, with a total of 81 control rules, and the rule weights are dynamically adjusted according to the aging characteristics of the cable insulation material; Step S74: Self-tuning of PID parameters, output quantity defuzzification formula: ; where is the activation degree, is the output value corresponding to the rule, and the proportional-integral-differential coefficients are updated in real time to ensure that the adjustment period < 500 ms; Step S75: Establish an initial threshold matrix according to the cable model ; where are the partial discharge threshold, harmonic distortion rate, and traveling wave amplitude threshold respectively, and an adaptive adjustment equation is constructed, and the specific formula is as follows: ; where is the reference threshold, is the proportionality coefficient, is the load rate deviation, is the integral coefficient, is the integral term of the temperature rise rate, is the differential coefficient, is the transformation rate of the temperature influence factor, restricting the correction amplitude within the range of ±40% to prevent overshoot; Step S76: Introduce the Sigmoid function for non - linear limiting to ensure that the threshold is always within the safe range and complete the dynamic calculation of the warning threshold.

[0043] Embodiment 2, refer to Figure 2 As shown, the present invention is a distribution network cable fault warning system based on multi - terminal synchronous traveling wave detection, which can be used to execute the method content of Embodiment 1 of the present invention, including: a distributed traveling wave sensing array, a quantum time synchronization network, an intelligent analysis center, and an anti - interference transmission system; The distributed traveling wave sensing array includes a multi - modal sensor cluster and an edge computing node; the multi - modal sensor cluster deploys an electromagnetic - ultrasonic composite sensor at the cable joint, integrating a high - frequency current transformer and a piezoelectric vibration sensor for synchronous acquisition of multi - physical quantities of traveling waves; the edge computing node has an FPGA pre - processing chip built into each terminal for wavelet packet threshold denoising and marking the arrival time of traveling waves; The quantum time synchronization network includes a Beidou / GPS - rubidium atomic clock dual - mode timing module and a PTPv2 protocol optimized channel; the Beidou / GPS - rubidium atomic clock dual - mode timing module is used for the time synchronization error of each terminal; the PTPv2 protocol optimized channel is used to establish a hybrid communication link based on the cable metal sheath to ensure that the time synchronization signal can still be transmitted when the communication is interrupted; The intelligent analysis center includes a feature fusion engine, a deep prediction model, and a dynamic risk assessment module; the feature fusion engine uses a multi - source data fusion algorithm with an attention mechanism to establish a time - domain - frequency - domain - energy - domain joint feature matrix; the deep prediction model is a spatio - temporal sequence prediction network based on TCN - GRU, embedding prior knowledge of cable topology for early fault feature extraction; the dynamic risk assessment module is used to construct a five - level warning index system by combining the D - S fusion algorithm of the improved evidence theory; The anti - interference transmission system is a dual - channel redundancy transmission of power line carrier and NB - IoT.

[0044] It should be noted that in the above system embodiments, the included units are only divided according to functional logic, but are not limited to the above division as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for easy distinction and do not limit the protection scope of the present invention.

[0045] In addition, those of ordinary skill in the art can understand that all or part of the steps in implementing the methods of the above embodiments can be completed by instructing relevant hardware through a program, and the corresponding program can be stored in a computer-readable storage medium.

[0046] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A distribution network cable fault early warning method based on multi-terminal synchronous traveling wave detection, characterized in that: The steps include: Step S1: deploy at least three synchronous detection terminals at each branch end of the cable, and trigger the sampling rate synchronously; Step S2: Establish a master-slave synchronization network, use FPGA to generate sampling clock to achieve multi-terminal synchronous startup, and perform spectrum correction on the collected data; Step S3: Establish a cable network traveling wave propagation matrix model and calculate the direction of arrival by using the cross-correlation time difference method; Step S4: constructing a real-time correction model based on distributed temperature sensor data; Step S5: extracting early fault features and constructing a three-dimensional feature tensor; Step S6: preliminary diagnosis of edge nodes and feature classification using transfer learning to optimize the network; Step S7: Combine the cable load rate with the ambient temperature and humidity to establish a fuzzy PID-based early warning threshold adaptive model.

2. A distribution network cable fault early warning method based on multi-terminal synchronous traveling wave detection according to claim 1, characterized in that: In step S1, the synchronous detection end is deployed at the beginning and end and the middle joint of the cable trunk line and the T-connected branch to form a multi-terminal monitoring array; the synchronous detection end includes a high-frequency current transformer and a piezoelectric vibration sensor; the high-frequency current transformer is used to collect the core wire current; the piezoelectric vibration sensor is used to shield the electromagnetic field of the layer and the mechanical vibration signal.

3. A distribution network cable fault early warning method based on multi-terminal synchronous traveling wave detection according to claim 1, characterized in that: In step S2, a master-slave synchronization network is established based on Beidou / GPS; wherein the master clock node periodically sends 1PPS second pulses, and the slave node transmits PTPv2 protocol messages through the cable sheath; the specific process of the spectrum correction is as follows: Step S21: using a parameterized Vincent window to perform signal truncation; Step S22: Perform 4096-point FFT on the windowed new signal to obtain an initial spectrum, and calculate the maximum amplitude point within the main lobe range; Step S23: constructing a cubic spline interpolation function within a specified interval to solve the true frequency deviation; Step S24: establishing an amplitude-phase correction model for iteration; Step S25: monitor the power frequency offset in real time and establish a dynamic compensation matrix; Step S26: construct a multi-filter architecture to perform anti-aliasing processing.

4. A distribution network cable fault early warning method based on multi-terminal synchronous traveling wave detection according to claim 1, characterized in that: In step S3, the specific process of establishing the cable network traveling wave propagation matrix model is as follows: Step S31: constructing a distribution network weighted graph according to the distribution network cable system; Step S32: Establish a time-varying propagation matrix to obtain the transmission function during the traveling wave transmission process; Step S33: establishing a joint differential equation for traveling wave propagation; Step S34: performing continuous wavelet transform on the two signals and then calculating the cross-correlation; Step S35: constructing an interpolation function within ±3 sampling points of the cross-correlation peak; Step S36: construct a cost function to distinguish direct waves from reflected waves, and calculate the direction of arrival.

5. A distribution network cable fault early warning method based on multi-terminal synchronous traveling wave detection according to claim 4, characterized in that: In step S36, the process of calculating the direction of arrival is as follows: Step S361: Calculate the time difference between each of the N terminals to form a time difference matrix; Step S362: performing path inversion using the propagation matrix; Step S363: construct an array flow matrix and solve the spatial spectrum peak.

6. A distribution network cable fault early warning method based on multi-terminal synchronous traveling wave detection according to claim 1, characterized in that: In step S5, the specific process of extracting early fault features and constructing a three-dimensional feature tensor is as follows: Step S51: dynamically selecting the optimal wavelet basis function according to the cable fault characteristic frequency band; Step S52: introducing a local-global weight factor and calculating the dynamic energy entropy ratio; Step S53: 6-layer wavelet packet decomposition, calculating the normalized WEE value for each node to form a 64-dimensional feature vector, and retaining the first 8 most discriminative features through KL divergence; Step S54: Calculate the time synchronization difference based on the arrival time differences of the traveling waves of multiple terminals; Step S55: Fusing electromagnetic / vibration dual-modal data to construct a hybrid spectrum matrix; Step S56: perform truncated SVD decomposition to form a three-dimensional feature tensor.

7. A distribution network cable fault early warning method based on multi-terminal synchronous traveling wave detection according to claim 1, characterized in that: In step S6, the specific process of the edge node preliminary diagnosis is as follows: Step S61: dynamically segment the real-time traveling wave signal using a sliding window mechanism, and the window length is adaptively adjusted according to the cable length; Step S62: extracting the third-order feature vectors of the segmented waveform: time domain features, frequency domain features and morphological features; Step S63: introducing a dynamic time adjustment optimization algorithm and adding an adaptive noise suppression layer; Step S64: Establish a fault waveform feature library and perform a three-level similarity determination.

8. A distribution network cable fault early warning method based on multi-terminal synchronous traveling wave detection according to claim 1, characterized in that: In step S6, when the transfer learning optimizes the network, the multi-terminal data received from the cloud are aligned in time and space, and the improved SVD noise reduction processing is performed to retain the first three principal components; multi-scale feature fusion is performed, and the triplet loss function is used to perform dynamic category weight adjustment.

9. A distribution network cable fault early warning method based on multi-terminal synchronous traveling wave detection according to claim 1, characterized in that: In step S7, the specific steps of establishing the early warning threshold adaptive model based on fuzzy PID are as follows: Step S71: measuring the cable core current in real time through the Rogowski coil and calculating the load rate; Step S72: deploy distributed temperature and humidity sensors, install one set for every 200-meter cable segment, and establish an environmental factor matrix; Step S73: define three fuzzy input quantities and establish a fuzzy rule base; Step S74: PID parameter self-tuning, real-time update of proportional-integral-differential coefficients; Step S75: establishing an initial threshold matrix according to the cable model and constructing an adaptive adjustment equation; Step S76: Introduce the Sigmoid function to perform nonlinear limiting and complete the dynamic calculation of the warning threshold.

10. A distribution network cable fault early warning system based on multi-terminal synchronous traveling wave detection, comprising a distributed traveling wave sensor array, a quantum time synchronization network, an intelligent analysis center and an anti-interference transmission system, characterized in that: The distributed traveling wave sensor array includes a multimodal sensor cluster and an edge computing node; the multimodal sensor cluster deploys an electromagnetic-ultrasonic composite sensor at the cable joint, integrates a high-frequency current transformer and a piezoelectric vibration sensor, and is used for synchronous acquisition of traveling wave multi-physical quantities; the edge computing node is a built-in FPGA pre-processing chip for each terminal, which is used for wavelet packet threshold denoising and traveling wave arrival time marking; The quantum time synchronization network includes a Beidou / GPS-rubidium atomic clock dual-mode timing module and a PTPv2 protocol optimization channel; the Beidou / GPS-rubidium atomic clock dual-mode timing module is used for the time synchronization error of each terminal; the PTPv2 protocol optimization channel is used to establish a hybrid communication link based on the cable metal sheath to ensure that the time synchronization signal can still be transmitted when the communication is interrupted; The intelligent analysis center includes a feature fusion engine, a deep prediction model and a dynamic risk assessment module; The feature fusion engine adopts a multi-source data fusion algorithm with an attention mechanism to establish a joint feature matrix of the time domain, frequency domain and energy domain; the deep prediction model is based on a TCN-GRU spatiotemporal sequence prediction network, embedding cable topology prior knowledge for early fault feature extraction; The dynamic risk assessment module is used to combine the DS fusion algorithm of the improved evidence theory to build a five-level early warning indicator system; The anti-interference transmission system is a power line carrier and NB-IoT dual-channel redundant transmission.

Images