Search

CN-121978478-A - Cable joint partial discharge fiber grating acoustic wave intelligent sensing and detecting method

CN121978478ACN 121978478 ACN121978478 ACN 121978478ACN-121978478-A

Abstract

The invention provides a cable joint partial discharge fiber bragg grating sound wave intelligent sensing and detecting method, which relates to the technical field of cable partial discharge detection and comprises the steps of obtaining multichannel time domain spectral response data of a tested cable joint through a fiber bragg grating sensing array and demodulating the multichannel time domain spectral response data to obtain a multidimensional feature vector group, decomposing a waveform signal into narrow-band eigenvalue components, extracting energy density through Hilbert transformation and executing singular value decomposition, screening a dominant modal component reconstruction waveform signal to obtain stress wave time sequence features, calculating a waveform high-order cross-correlation tensor as a superside weight, extracting a time-frequency amplitude feature vector as an initial node feature, constructing a space-time supergraph and generating a node embedding vector through graph convolution, initializing a Gaussian noise field based on the node embedding vector and setting condition information, executing reverse sampling denoising to generate a three-dimensional space probability density field to identify initial positioning coordinates, and iteratively correcting the positioning coordinates and wave speed values until residual errors are converged.

Inventors

  • ZHU SHI
  • NI LIANG
  • YAN ZHIXUE

Assignees

  • 天津欧利信科技有限公司

Dates

Publication Date
20260505
Application Date
20260127

Claims (10)

  1. 1. The intelligent sensing and detecting method for the partial discharge fiber grating acoustic wave of the cable joint is characterized by comprising the following steps: Acquiring multichannel time domain spectral response data corresponding to a tested cable joint through a fiber bragg grating sensing array, and demodulating the multichannel time domain spectral response data based on a preset sensitivity coefficient to obtain a multidimensional feature vector group; Decomposing waveform signals in the multidimensional feature vector group into narrow-band eigenmode components, extracting energy density through Hilbert transformation to construct an energy density matrix, executing singular value decomposition on the energy density matrix, and screening a reconstruction waveform signal of a dominant mode component to obtain stress wave time sequence characteristics; Calculating a waveform high-order cross-correlation tensor of stress wave time sequence characteristics as an over-edge weight, taking a time-frequency amplitude characteristic vector in the stress wave time sequence characteristics as an initial node characteristic, constructing a space-time over-graph based on the over-edge weight and the initial node characteristic, generating an over-edge representation by graph convolution aggregation of the initial node characteristic in the space-time over-graph, aggregating the over-edge representation and obtaining a node embedding vector by step neighborhood aggregation; And initializing a Gaussian noise field based on the node embedded vector, setting condition information, performing inverse sampling denoising based on the condition information to generate an initial three-dimensional space probability density field, identifying a local maximum point as an initial positioning coordinate, calculating theoretical propagation time based on a preset wave velocity value, calculating a residual error by combining the stress wave time sequence characteristic, and iteratively correcting the initial positioning coordinate and the wave velocity value until the root mean square of the residual error is smaller than a preset residual error threshold value to obtain a corrected positioning coordinate.
  2. 2. The method of claim 1, wherein obtaining multichannel time-domain spectral response data corresponding to the tested cable joint through the fiber bragg grating sensing array and demodulating based on a preset sensitivity coefficient to obtain the multidimensional feature vector group comprises: Collecting grating periodic variation caused by stress waves generated by partial discharge through fiber bragg grating sensing arrays uniformly distributed along the circumferential direction of the cable joint to be tested, and recording the variation curve of Bragg wavelength drift quantity of each grating node along with time in real time to obtain multichannel time domain spectral response data; Performing baseline drift correction on the multichannel time domain spectrum response data, identifying wavelength mutation points exceeding standard deviation multiples of background noise in each channel as effective signal starting moments, and extracting wavelength drift sequences in an effective signal time window; Converting wavelength drift amount of each sampling point in the wavelength drift sequence into stress wave amplitude values through the preset sensitivity coefficient, constructing stress wave time domain waveform signals corresponding to each grating node, and forming a time-space associated waveform signal group by combining space coordinate information of each grating node; And performing peak detection on the time domain waveforms of all channels in the waveform signal group to extract peak amplitude, determining the arrival time difference of all channels relative to a reference channel through cross-correlation analysis, calculating the arrival time of all channels by combining the absolute arrival time of the reference channel, performing short-time Fourier transform on the time domain waveforms of all channels to obtain a time spectrum, calculating a spectrum centroid, and combining the waveform signal group, the arrival time, the peak amplitude and the spectrum centroid into a multidimensional feature vector group.
  3. 3. The method of claim 1, wherein decomposing the waveform signals in the multi-dimensional feature vector set into narrowband eigenmode components and extracting energy densities through hilbert transformation to construct an energy density matrix, performing singular value decomposition on the energy density matrix and screening dominant mode component reconstructed waveform signals to obtain stress wave timing features comprises: initializing candidate modal components corresponding to waveform signals in the multidimensional feature vector group, minimizing spectrum bandwidth to obtain a preliminary decomposition result, determining frequency tracks based on the preliminary decomposition result, calculating the similarity of the frequency tracks among different channels, determining similar modal pairs by combining a preset similarity threshold, and performing cross-correlation analysis and time compensation on the similar modal pairs to obtain narrow-band intrinsic modal components; Performing Hilbert transformation on the narrow-band eigenmode components to obtain analytic signals, extracting instantaneous amplitude values as instantaneous energy densities, arranging the instantaneous energy densities according to time sampling points to form a time sequence, and organizing the time sequence to obtain an energy density matrix; Performing singular value decomposition on the energy density matrix to obtain a singular value sequence and a corresponding left vector matrix and right vector matrix, extracting a time mode based on the left vector matrix and calculating to obtain a time metric, extracting a space mode based on the right vector matrix and calculating to obtain a space metric, calculating a smoothness score according to the time metric and the space metric and determining a screening index, screening the singular value sequence based on the screening index to determine dominant singular values and reconstructing to obtain a noise reduction energy density matrix, performing inverse Hilbert transform on the noise reduction energy density matrix to obtain a time domain waveform, and summing and superposing the time domain waveform and extracting a time-frequency amplitude characteristic parameter to obtain a stress wave time sequence characteristic.
  4. 4. The method of claim 1, wherein calculating a waveform high-order cross-correlation tensor of stress wave timing features as a superside weight, wherein using a time-frequency amplitude feature vector in stress wave timing features as an initial node feature, and wherein constructing a spatio-temporal supergraph based on the superside weight and the initial node feature comprises: dividing a time-frequency amplitude characteristic vector in the stress wave time sequence characteristic into a plurality of time segments according to a time window, respectively constructing time-frequency amplitude characteristic vectors in each time segment into graph nodes, randomly selecting three graph node combinations in the graph nodes of the current time segment to obtain a candidate superside node group, executing third-order tensor outer product operation on the time-frequency amplitude characteristic vector corresponding to the candidate superside node group, extracting tensor norms as cooperative strength, and normalizing the cooperative strength to obtain the superside weight; Extracting time domain peak amplitude, dominant frequency and energy gravity center time positions in the stress wave time sequence characteristics and splicing to form initial node characteristics of the graph nodes; And constructing a superside incidence matrix according to the superside weight, constructing a node feature matrix according to the initial node feature vector, multiplying the transpose of the superside incidence matrix by the superside incidence matrix to obtain a node indirect matrix, multiplying the node indirect matrix by the node feature matrix and performing nonlinear transformation to obtain updated node features, splicing the superside incidence matrix corresponding to each time segment along the time dimension to obtain a space-time superside matrix, and constructing a space-time supergraph by combining the updated node features.
  5. 5. The method of claim 1, wherein generating a hyperedge representation in the spatiotemporal hypergraph by graph convolution aggregating the initial node features, aggregating the hyperedge representation and obtaining a node embedding vector by rank neighborhood aggregation comprises: Obtaining graph nodes in the space-time hypergraph, calculating a characteristic covariance matrix of initial node characteristics corresponding to each graph node, extracting a principal component direction as an internal manifold base of the hyperedge, projecting the initial node characteristics to the internal manifold base to obtain manifold coordinate representation, calculating manifold distances among the graph nodes, constructing adaptive adjacent weights in the hyperedge based on the manifold distances, carrying out manifold perception weighted aggregation on the initial node characteristics, and carrying out multiplication gating in combination with the hyperedge weights to obtain the hyperedge representation; Extracting an associated superside set corresponding to each graph node, calculating mutual information between superside representations corresponding to each associated superside and superside representations corresponding to the current graph node, and screening the superside representations in the associated superside set based on the mutual information to obtain first-level node characteristics; And constructing a Laplace matrix based on the first-level node characteristics, extracting a low-frequency characteristic vector through spectrum decomposition to obtain a spectrum filter, carrying out frequency domain filtering and polynomial expansion on the first-level node characteristics corresponding to each graph node through the spectrum filter to obtain a multi-order neighborhood aggregation coefficient, and carrying out weighted aggregation on the superside representation based on the multi-order neighborhood aggregation coefficient to obtain the node embedding vector.
  6. 6. The method of claim 1, wherein initializing a gaussian noise field based on the node embedded vector and setting condition information, performing inverse sampling denoising based on the condition information to generate an initial three-dimensional spatial probability density field and identifying local maxima points as initial positioning coordinates comprises: Performing feature coding and space broadcasting operation on the node embedded vector to obtain space condition features, initializing a three-dimensional Gaussian noise field conforming to standard normal distribution, splicing the three-dimensional Gaussian noise field with the corresponding space condition features to obtain a conditional noise field, setting a total time step, initializing a current time step, performing convolution change operation on the conditional noise field and time codes corresponding to the current time step to obtain a noise component, subtracting the noise component from the conditional noise field representation, adding a preset scaling factor to obtain a denoising field representation, taking the denoising field representation as the conditional noise field of the next time step, decrementing the time step, and repeatedly executing until the time step returns to zero to obtain an initial three-dimensional space probability density field; And screening grid points in the initial three-dimensional space probability density field based on a preset probability density threshold value to obtain an initial grid point set, identifying initial grid points with probability density values larger than those of all adjacent grid points in the initial grid point set as candidate maximum value points, calculating gradient amplitude values between the probability density values of the candidate maximum value points and the probability density values of the corresponding adjacent grid points, arranging the candidate maximum value points in a descending order according to the gradient amplitude values, and selecting a first candidate maximum value point to extract coordinates in a three-dimensional space to obtain initial positioning coordinates.
  7. 7. The method of claim 1, wherein calculating theoretical propagation time based on a preset wave velocity value and calculating a residual in combination with the stress wave timing feature, iteratively correcting the initial positioning coordinates and the wave velocity value until a root mean square of the residual is less than a preset residual threshold value to obtain corrected positioning coordinates comprises: Calculating the space distance between the initial positioning coordinates and each sensor position, calculating theoretical propagation time by combining a preset wave speed value, extracting actual arrival time from the stress wave time sequence characteristics, and calculating the time difference between the theoretical propagation time and the actual arrival time as a time residual; Constructing a state vector containing a three-dimensional coordinate component of the initial positioning coordinate and the preset wave velocity value, judging whether the time residual is smaller than a preset time threshold, and outputting the initial positioning coordinate in the state vector as a corrected positioning coordinate if the time residual is smaller than the preset time threshold; If not, applying random disturbance to the state vector and recalculating the time residual to obtain a jacobian matrix, calculating a gradient vector based on the transpose of the jacobian matrix and the time residual, constructing a damping matrix based on the jacobian matrix and a preset damping coefficient, constructing a linear equation set based on the damping matrix and the gradient vector and solving to obtain a correction increment, updating the state vector based on the correction increment to obtain an updated state vector, calculating the root mean square of the current time residual based on the updated state vector as an updated residual root mean square value and correcting the damping coefficient, repeating iteration until the updated residual root mean square value is smaller than a preset residual threshold value, and obtaining and outputting a corrected positioning coordinate.
  8. 8. An intelligent sensing and detecting system for cable joint partial discharge fiber grating acoustic wave, which is used for realizing the method as set forth in any one of the preceding claims 1-7, and is characterized by comprising: the first unit is used for acquiring multichannel time domain spectrum response data corresponding to the tested cable joint through the fiber bragg grating sensing array and demodulating the multichannel time domain spectrum response data based on a preset sensitivity coefficient to obtain a multidimensional feature vector group; the second unit is used for decomposing the waveform signals in the multidimensional feature vector group into narrow-band intrinsic mode components, extracting energy density through Hilbert transformation to construct an energy density matrix, executing singular value decomposition on the energy density matrix, and screening a dominant mode component reconstruction waveform signal to obtain stress wave time sequence characteristics; The third unit is used for calculating waveform high-order cross-correlation tensor of stress wave time sequence characteristics as an over-edge weight, taking a time-frequency amplitude characteristic vector in the stress wave time sequence characteristics as an initial node characteristic, constructing a space-time hypergraph based on the over-edge weight and the initial node characteristic, generating an over-edge representation by integrating the initial node characteristic through graph convolution in the space-time hypergraph, integrating the over-edge representation and obtaining a node embedding vector through order neighborhood integration; and a fourth unit, configured to initialize a gaussian noise field based on the node embedding vector and set condition information, perform inverse sampling denoising based on the condition information to generate an initial three-dimensional space probability density field and identify a local maximum point as an initial positioning coordinate, calculate a theoretical propagation time based on a preset wave velocity value and calculate a residual error in combination with the stress wave time sequence feature, and iteratively correct the initial positioning coordinate and the wave velocity value until the root mean square of the residual error is smaller than a preset residual error threshold value to obtain a corrected positioning coordinate.
  9. 9. An electronic device, comprising: A processor; A memory for storing processor-executable instructions; Wherein the processor is configured to invoke the instructions stored in the memory to perform the method of any of claims 1 to 7.
  10. 10. A computer readable storage medium having stored thereon computer program instructions, which when executed by a processor, implement the method of any of claims 1 to 7.

Description

Cable joint partial discharge fiber grating acoustic wave intelligent sensing and detecting method Technical Field The invention relates to the technical field of cable partial discharge detection, in particular to an intelligent sensing and detecting method for a cable joint partial discharge fiber bragg grating acoustic wave. Background The power cable is used as an important component of a power system, is directly related to the safety and stability of a power grid, and the cable joint is used as the weakest link in a cable line, so that insulation defects are easy to occur due to the influence of factors such as a manufacturing process, installation quality, operation environment and the like, and further partial discharge phenomenon is generated. Partial discharge is a main sign of cable insulation degradation, and long-term partial discharge gradually erodes insulation materials, and finally causes insulation breakdown faults. Therefore, the cable connector partial discharge is effectively monitored and accurately positioned, and the method has important significance for preventing cable faults and guaranteeing the safe operation of the power grid. The existing cable partial discharge detection technology mainly comprises an electrical detection method, an ultrasonic detection method, an ultrahigh frequency detection method and the like, and the fiber bragg grating sensing technology is used as a novel acoustic wave detection means, has the advantages of being intrinsically safe, resistant to electromagnetic interference, capable of realizing distributed measurement and the like, and is gradually applied to the field of cable partial discharge monitoring. However, the existing cable joint partial discharge fiber grating acoustic wave detection and positioning technology still has the problems that the suppression capability for complex background noise and multi-source interference is insufficient, the characteristic expression is insufficient due to the fact that multi-channel space-time associated information is not fully excavated, the adaptability to wave speed uncertainty and propagation path complexity is weak, the positioning error is large, and engineering requirements for accurate positioning of the cable joint partial discharge are difficult to meet. Disclosure of Invention The embodiment of the invention provides a cable joint partial discharge fiber grating acoustic wave intelligent sensing and detecting method, which at least can solve part of problems in the prior art. In a first aspect of the embodiment of the present invention, a method for intelligently sensing and detecting a cable connector partial discharge fiber bragg grating acoustic wave is provided, including: Acquiring multichannel time domain spectral response data corresponding to a tested cable joint through a fiber bragg grating sensing array, and demodulating the multichannel time domain spectral response data based on a preset sensitivity coefficient to obtain a multidimensional feature vector group; Decomposing waveform signals in the multidimensional feature vector group into narrow-band eigenmode components, extracting energy density through Hilbert transformation to construct an energy density matrix, executing singular value decomposition on the energy density matrix, and screening a reconstruction waveform signal of a dominant mode component to obtain stress wave time sequence characteristics; Calculating a waveform high-order cross-correlation tensor of stress wave time sequence characteristics as an over-edge weight, taking a time-frequency amplitude characteristic vector in the stress wave time sequence characteristics as an initial node characteristic, constructing a space-time over-graph based on the over-edge weight and the initial node characteristic, generating an over-edge representation by graph convolution aggregation of the initial node characteristic in the space-time over-graph, aggregating the over-edge representation and obtaining a node embedding vector by step neighborhood aggregation; And initializing a Gaussian noise field based on the node embedded vector, setting condition information, performing inverse sampling denoising based on the condition information to generate an initial three-dimensional space probability density field, identifying a local maximum point as an initial positioning coordinate, calculating theoretical propagation time based on a preset wave velocity value, calculating a residual error by combining the stress wave time sequence characteristic, and iteratively correcting the initial positioning coordinate and the wave velocity value until the root mean square of the residual error is smaller than a preset residual error threshold value to obtain a corrected positioning coordinate. In an alternative embodiment of the present invention, The method for obtaining the multi-channel time domain spectral response data corresponding to the tested cable connector through the fiber