Cable fault acoustic print monitoring and early warning method and system based on feature domain reconstruction

Abstract

This invention relates to the field of fault detection technology, specifically disclosing a method and system for cable fault acoustic signature monitoring and early warning based on feature domain reconstruction. The method includes: acquiring acoustic signature signals generated during the operation of the target cable and performing data preprocessing; selecting a reference position and a acquisition position to be corrected based on the signal-to-noise ratio of the preprocessed acoustic signature signals; performing spatial feature correction on the preprocessed acoustic signature signals corresponding to the acquisition position to be corrected; extracting features from both the preprocessed and corrected acoustic signature signals corresponding to the reference position; performing weighted fusion of a reference acoustic signature feature parameter set and the corrected acoustic signature feature parameter set; matching the fused acoustic signature feature parameter set with a preset fault acoustic signature feature template library; generating graded early warning information based on the matching results; and outputting the graded early warning information to a monitoring terminal. This invention can effectively improve the accuracy and reliability of cable fault monitoring.

Description

Technical Field

[0001] This invention relates to the field of fault detection technology, and in particular to a method and system for monitoring and early warning of cable fault acoustic signatures based on feature domain reconstruction. Background Technology

[0002] As a crucial component of power transmission, the real-time monitoring of cable operating status is of paramount importance for ensuring the safe and stable operation of the power grid. When cables experience faults such as insulation aging, short circuits, or mechanical damage, they generate acoustic signature signals with specific frequency and time-domain characteristics. The acoustic signature characteristics differ significantly depending on the type of fault. Therefore, by collecting and analyzing the acoustic signature signals of cables during operation, effective identification and early warning of cable fault types can be achieved. Existing technologies employ multiple sensors to collect cable acoustic signature signals and utilize methods such as Mel-frequency cepstral coefficients for feature extraction, combined with support vector machines or deep learning network models to achieve automatic fault type identification.

[0003] However, existing technologies still have the following shortcomings. First, existing solutions lack a systematic spatial correlation between the deployment of multiple sensors and signal processing. The signals acquired by each sensor are typically processed as independent data sources, lacking a technical solution to quantify and correlate the spatial layout coordinates of the sensors with the differences in the propagation path of the acoustic signature signal for signal correction. Second, existing technologies lack a mechanism for actively compensating for signal distortion caused by different propagation paths—when the acoustic signature signal propagates from the fault point to sensors at different locations, differences in propagation distance and medium result in different arrival time delays and frequency domain amplitude attenuation. Existing methods do not systematically quantify and correct this physical difference, leading to a lack of consistency in the time and frequency domains of signals acquired by sensors at different locations. Third, after signal acquisition and feature extraction, existing solutions often determine the fusion weights based on fixed empirical values ​​or simple signal quality indicators, lacking a dynamic evaluation and adaptive adjustment mechanism based on the correction effect. These deficiencies prevent optimal integration of signals acquired by multiple sensors during the fusion process, affecting the accuracy and reliability of fault identification. Therefore, there is an urgent need to develop a cable fault acoustic signature monitoring and early warning method that can perform spatial feature correction on multi-sensor acoustic signature signals and realize adaptive weighted fusion based on the correction effect, so as to solve the problems of insufficient signal distortion compensation and lack of dynamic fusion strategy in the existing technology, and improve the accuracy and reliability of cable fault monitoring. Summary of the Invention

[0004] This invention provides a method and system for monitoring and early warning of cable fault acoustic signatures based on feature domain reconstruction, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a cable fault acoustic signature monitoring and early warning method based on feature domain reconstruction, comprising: S01: Collect the acoustic fingerprint signal generated during the operation of the target cable, perform data preprocessing, and select the reference position and the collection position to be calibrated of the target cable based on the signal-to-noise ratio of the preprocessed acoustic fingerprint signal. S02: Based on the spatial layout coordinates of the acoustic sensor, the preprocessed acoustic signal corresponding to the reference position, and the preprocessed acoustic signal corresponding to the acquisition position to be corrected, a propagation path difference matrix is ​​constructed. Based on the propagation path difference matrix, spatial feature correction is performed on the preprocessed acoustic signal corresponding to the acquisition position to be corrected. S03: Perform feature extraction on the preprocessed voiceprint signal and the corrected voiceprint signal corresponding to the reference position respectively to obtain the reference voiceprint feature parameter set and the corrected voiceprint feature parameter set; S04: Perform weighted fusion of the reference voiceprint feature parameter set and the corrected voiceprint feature parameter set to obtain a fused voiceprint feature parameter set; S05: Match the fused voiceprint feature parameter set with the preset fault voiceprint feature template library, generate graded early warning information based on the matching result, and output the graded early warning information to the monitoring terminal.

[0006] In a preferred embodiment, the process of acquiring acoustic signature signals generated during the operation of the target cable and performing data preprocessing, and selecting the reference position and the acquisition position to be calibrated of the target cable based on the signal-to-noise ratio of the preprocessed acoustic signature signals, includes: Based on pre-deployed acoustic sensors, acoustic signals generated during the operation of the target cable are collected; The voiceprint signal is amplified, and the amplified analog voiceprint signal is converted from analog to digital to obtain a preprocessed voiceprint signal. Noise power estimation and signal power estimation are performed on the preprocessed acoustic signal respectively, and the signal-to-noise ratio corresponding to each acquisition location is calculated. The acquisition position corresponding to the highest signal-to-noise ratio was selected as the reference position; The remaining acquisition positions, excluding the reference position, are arranged in descending order of signal-to-noise ratio and marked as acquisition positions to be calibrated.

[0007] In a preferred embodiment, the step of constructing a propagation path difference matrix based on the spatial layout coordinates of the voiceprint sensor, the preprocessed voiceprint signal corresponding to the reference position, and the preprocessed voiceprint signal corresponding to the acquisition position to be calibrated, and performing spatial feature correction on the preprocessed voiceprint signal corresponding to the acquisition position to be calibrated based on the propagation path difference matrix, includes: The spatial distance between the reference position and the acquisition position to be corrected is measured to obtain the spatial distance parameter between the acquisition position to be corrected and the reference position. Time-frequency transformation is performed on the preprocessed acoustic fingerprint signal corresponding to the reference position and the preprocessed acoustic fingerprint signal corresponding to the acquisition position to be corrected, respectively. Extract the reference time-domain waveform and reference frequency-domain amplitude spectrum, as well as the time-domain waveform to be calibrated and the frequency-domain amplitude spectrum to be corrected from the time-frequency transformation results; Perform cross-correlation calculation on the time-domain waveform to be calibrated and the reference time-domain waveform to obtain the arrival time difference of the acquisition position to be calibrated relative to the reference position; The amplitude ratio between the amplitude spectrum of the frequency domain to be corrected and the amplitude spectrum of the reference frequency domain is calculated to obtain the relative amplitude difference coefficient between the acquisition position to be corrected and the reference position. Based on the spatial layout order of the acquisition locations to be corrected, the spatial distance parameters, the arrival time difference, and the relative amplitude difference coefficient are arranged in a matrix.

[0008] In a preferred embodiment, the step of performing spatial feature correction on the preprocessed acoustic signature signal corresponding to the acquisition location to be corrected based on the propagation path difference matrix includes: Extract the arrival time difference corresponding to the acquisition position to be corrected from the propagation path difference matrix, and determine the time delay compensation direction corresponding to the acquisition position to be corrected based on the sign of the arrival time difference. Based on the time delay compensation direction, the preprocessed voiceprint signal corresponding to the acquisition position to be corrected is time-aligned. Extract the relative amplitude difference coefficients corresponding to the acquisition positions to be corrected from the propagation path difference matrix, and perform a reciprocal operation on the relative amplitude difference coefficients to obtain the amplitude compensation gain; Based on the amplitude compensation gain, the amplitude of the preprocessed acoustic signal corresponding to the acquisition position to be corrected is adjusted so that the amplitude value of the preprocessed acoustic signal at each frequency point is consistent with the preprocessed acoustic signal corresponding to the reference position.

[0009] In a preferred embodiment, the step of extracting features from the preprocessed voiceprint signal and the corrected voiceprint signal corresponding to the reference position to obtain a reference voiceprint feature parameter set and a corrected voiceprint feature parameter set includes: The Mel frequency cepstral coefficient analysis method was used to extract features from the preprocessed and corrected voiceprint signals corresponding to the reference positions, respectively. Based on the preset number of sampling points and the preset frame shift step size, the preprocessed voiceprint signal and the corrected voiceprint signal corresponding to the reference position are processed into frames to obtain a voiceprint frame sequence. The signal of each frame in the voiceprint frame sequence is pre-emphasized to compensate for the high-frequency energy loss of the signal during propagation, and a pre-emphasized signal is obtained. A window function is applied to the pre-emphasized signal to obtain a windowed signal; Perform a Fast Fourier Transform on the windowed signal to obtain the frequency domain amplitude spectrum; The power spectral density estimate is obtained by squaring the frequency domain amplitude spectrum.

[0010] In a preferred embodiment, after obtaining the power spectral density estimate, the method further includes: The power spectral density estimate is input into the Mel filter bank to obtain the bandgap energy value; Taking the natural logarithm of the frequency band energy values ​​yields logarithmic energy spectrum data; Perform a discrete cosine transform on the logarithmic energy spectrum data, extract the low-order discrete cosine transform coefficients, and obtain the Mel frequency cepstral coefficient eigenvector. The Mel frequency cepstral coefficient feature vectors are aggregated in chronological order to obtain a reference voiceprint feature parameter set and a corrected voiceprint feature parameter set.

[0011] In a preferred embodiment, after obtaining the frequency band energy value, the method further includes: Based on the frequency band energy value, the cosine similarity between the frequency band energy distribution of the corrected voiceprint signal and the frequency band energy distribution of the preprocessed voiceprint signal corresponding to the reference position is calculated to obtain the correction consistency index. The formula for calculating the corrected consistency index is as follows: ; in, The number of filters in a Mel filter bank. For the first The frequency band energy value of the corrected acoustic signature signal at the passband output of each filter. For the first The frequency band energy value of the preprocessed acoustic signature signal corresponding to the reference position of the passband output of each filter. The corrected consistency index is denoted as .

[0012] In a preferred embodiment, the weighted fusion of the reference voiceprint feature parameter set and the corrected voiceprint feature parameter set to obtain a fused voiceprint feature parameter set includes: Calculate the calibration consistency index of the data acquisition locations to be calibrated; The ratio of the correction consistency index corresponding to the acquisition location to be corrected to the sum of the correction consistency indices is used as the dynamic fusion weight. Extract the corrected voiceprint feature vector from the set of corrected voiceprint feature parameters, and multiply the dynamic fusion weight with the corrected voiceprint feature vector to obtain the weighted corrected voiceprint feature vector. The reference voiceprint feature vector is extracted from the reference voiceprint feature parameter set, and the reference voiceprint feature vector and the weighted and corrected voiceprint feature vector are summed to obtain the fused voiceprint feature parameter set.

[0013] In a preferred embodiment, the step of matching the fused voiceprint feature parameter set with a preset fault voiceprint feature template library, generating graded early warning information based on the matching result, and outputting the graded early warning information to the monitoring terminal includes: The feature distance between the fused voiceprint feature parameter set and the standard voiceprint feature parameter templates corresponding to each fault type in the preset fault voiceprint feature template library is calculated. The fault type with the smallest feature distance is selected as the current fault type of the target cable, and the minimum feature distance is normalized and mapped to obtain the fault confidence. Based on the fault confidence level, the warning level is determined, the fault type, the fault confidence level, and the warning level are encapsulated into graded warning information, and the graded warning information is output to the monitoring terminal.

[0014] To address the above problems, the present invention also provides a cable fault acoustic signature monitoring and early warning system based on feature domain reconstruction, the system comprising: The signal acquisition module is used to acquire acoustic fingerprint signals generated during the operation of the target cable, perform data preprocessing, and select the reference position and the acquisition position to be calibrated of the target cable based on the signal-to-noise ratio of the preprocessed acoustic fingerprint signal. The spatial correction module is used to construct a propagation path difference matrix based on the spatial layout coordinates of the acoustic sensor, the preprocessed acoustic signal corresponding to the reference position, and the preprocessed acoustic signal corresponding to the acquisition position to be corrected, and to perform spatial feature correction on the preprocessed acoustic signal corresponding to the acquisition position to be corrected based on the propagation path difference matrix. The feature extraction module is used to extract features from the preprocessed voiceprint signal and the corrected voiceprint signal corresponding to the reference position, respectively, to obtain a reference voiceprint feature parameter set and a corrected voiceprint feature parameter set; The weighted fusion module is used to perform weighted fusion on the reference voiceprint feature parameter set and the corrected voiceprint feature parameter set to obtain a fused voiceprint feature parameter set. The early warning output module is used to match the fused voiceprint feature parameter set with the preset fault voiceprint feature template library, generate graded early warning information based on the matching result, and output the graded early warning information to the monitoring terminal.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention constructs a propagation path difference matrix, systematically quantifying the arrival time difference and relative amplitude difference coefficient of each acquisition location to be corrected relative to a reference location. Based on this, it performs time delay compensation and frequency domain amplitude compensation on the preprocessed acoustic signature signal of the acquisition location to be corrected, achieving active correction of signal distortion caused by different propagation paths. In existing technologies, there is a lack of systematic spatial correlation between the deployment of multiple sensors and signal processing. The acquired signals of each sensor are usually processed as independent data sources, lacking a technical solution to quantify and correlate the spatial layout coordinates of the sensors with the differences in the propagation path of the acoustic signature signal for signal correction. This invention systematically fuses the spatial layout coordinates of the acoustic signature sensor with the time-frequency characteristics of the signal, enabling the acoustic signature signal of each acquisition location to achieve spatial consistency in both the time and frequency domains. This effectively eliminates the arrival time delay and frequency domain amplitude attenuation caused by propagation distance and medium differences, improving data quality from the signal source. Meanwhile, this invention introduces a correction consistency index into the Mel frequency cepstral coefficient feature extraction process. By calculating the cosine similarity between the frequency band energy distribution of the corrected voiceprint signal and the reference position signal, the correction effect is quantified into a measurable numerical index, filling the technical gap in the prior art where there is a lack of quantitative evaluation methods for correction effect.

[0016] 2. This invention implements an adaptive dynamic weighted fusion strategy based on correction effects. Existing schemes, after signal acquisition and feature extraction, often determine fusion weights based on fixed empirical values ​​or simple signal quality indicators, lacking a dynamic evaluation and adaptive adjustment mechanism based on correction effects. This invention normalizes the correction consistency index of each acquisition location to be corrected into dynamic fusion weights, ensuring that channels with better correction effects receive a greater contribution in the fusion, while effectively suppressing the contributions of channels with poorer correction effects. This mechanism achieves adaptive optimization of the fusion strategy, resulting in a higher signal-to-noise ratio and more representative fault features in the final fused acoustic signature feature parameter set used for fault matching. Furthermore, this invention uses a pre-defined fault acoustic signature feature template library for fault type matching, eliminating the need for large-scale training data and expensive computing resources required for deep learning models, thus reducing system deployment costs. Through the comprehensive application of the above techniques, this invention effectively improves the accuracy and reliability of cable fault acoustic signature monitoring, reduces false alarm and false negative rates, and provides an efficient and reliable technical solution for cable operation status monitoring. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a cable fault acoustic signature monitoring and early warning method based on feature domain reconstruction, provided in an embodiment of the present invention. Figure 2 This is a functional block diagram of a cable fault acoustic signature monitoring and early warning system based on feature domain reconstruction, provided in an embodiment of the present invention.

[0019] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0021] This application provides a method for monitoring and early warning of cable fault acoustic signatures based on feature domain reconstruction. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cluster of cloud servers. The server can be an independent server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0022] Reference Figure 1 The diagram shown is a flowchart illustrating a cable fault acoustic signature monitoring and early warning method based on feature domain reconstruction according to an embodiment of the present invention. In this embodiment, the cable fault acoustic signature monitoring and early warning method based on feature domain reconstruction includes: S01: Collect the acoustic fingerprint signal generated during the operation of the target cable, perform data preprocessing, and select the reference position and the collection position to be calibrated of the target cable based on the signal-to-noise ratio of the preprocessed acoustic fingerprint signal. In this embodiment of the invention, the step of acquiring acoustic signature signals generated during the operation of the target cable, performing data preprocessing, and selecting the reference position and the acquisition position to be calibrated of the target cable based on the signal-to-noise ratio of the preprocessed acoustic signature signals includes: Based on pre-deployed acoustic sensors, acoustic signals generated during the operation of the target cable are collected; The voiceprint signal is amplified, and the amplified analog voiceprint signal is converted from analog to digital to obtain a preprocessed voiceprint signal. Noise power estimation and signal power estimation are performed on the preprocessed acoustic signal respectively, and the signal-to-noise ratio corresponding to each acquisition location is calculated. The acquisition position corresponding to the highest signal-to-noise ratio was selected as the reference position; The remaining acquisition positions, excluding the reference position, are arranged in descending order of signal-to-noise ratio and marked as acquisition positions to be calibrated.

[0023] Multiple collection points are selected along the target cable laying path. Acoustic sensors are fixedly installed at each collection point to ensure that each sensor is in close contact with the outer surface of the cable to effectively receive acoustic signals. At the same time, the spatial layout coordinates of the sensors are recorded at each collection point.

[0024] Each acoustic sensor starts collecting data synchronously with the same sampling frequency and the same sampling duration. Each sensor converts the sensed acoustic vibrations into analog electrical signals, and each sensor outputs an analog acoustic signal corresponding to the collection location.

[0025] The analog voiceprint signals from each acquisition location are sent to the corresponding signal amplification circuit. The signal amplification circuit uses differential amplification to proportionally amplify the voltage amplitude of the analog voiceprint signal, so that the voltage range of the amplified analog voiceprint signal reaches the rated voltage range of the analog-to-digital converter input terminal.

[0026] The amplified analog voiceprint signal is input into the analog-to-digital converter corresponding to each acquisition location. The analog-to-digital converter samples the amplified analog voiceprint signal at equal time intervals with a fixed sampling frequency. The analog voltage value at each sampling moment is quantized into a digital quantity. The digital quantities at each sampling moment are arranged in chronological order to form the preprocessed voiceprint signal sequence corresponding to each acquisition location.

[0027] Noise power estimation is performed on the preprocessed acoustic signal sequence corresponding to each acquisition location. The preprocessed acoustic signal sequence is divided into multiple short time segments of equal length. The amplitude of each sampling point in each short time segment is squared, summed, and then divided by the total number of sampling points in the short time segment to obtain the short time energy value of each short time segment. The minimum value among all short time energy values ​​is selected as the estimated environmental noise power value of the acquisition location.

[0028] Signal power estimation is performed on the preprocessed voiceprint signal sequence corresponding to each acquisition location. The absolute value of the amplitude of each sampling point in the preprocessed voiceprint signal sequence is detected. The amplitude corresponding to the sampling point with the largest absolute value of amplitude is selected as the peak amplitude of the signal. Half of the square of the peak amplitude of the signal is used as the signal power estimate of the acquisition location.

[0029] The signal power estimate at each acquisition location is divided by its corresponding environmental noise power estimate to obtain the signal-to-noise ratio (SNR) value for each acquisition location.

[0030] The signal-to-noise ratio (SNR) values ​​of all acquisition locations are compared, and the acquisition location with the highest SNR value is selected as the reference location. The preprocessed voiceprint signal sequence corresponding to this reference location is used as the reference signal for subsequent steps.

[0031] The remaining acquisition positions, excluding the reference position, are arranged in descending order of their respective confidence-to-noise ratio values. All of these acquisition positions are marked as acquisition positions to be calibrated, and the preprocessed acoustic signature signal sequence corresponding to each acquisition position to be calibrated is used as the calibration signal for subsequent steps.

[0032] The beneficial effects are as follows: by separately estimating the noise power and signal power of the preprocessed acoustic signature signals at each acquisition location, the signal-to-noise ratio (SNR) of each acquisition location is quantitatively calculated, and a reference location is automatically selected based on the SNR value, providing objective data support for the signal quality of the reference location. Compared to the existing technology where the reference sensor location is selected based on manual experience, the SNR, as an objective indicator of signal quality, does not rely on the operator's experience judgment during its calculation. In actual working conditions where noise levels vary at different acquisition points along the cable, it can adaptively select the acquisition location with the optimal signal quality as the reference, providing a high-quality reference for the subsequent construction of the propagation path difference matrix and spatial feature correction. This avoids the problem of amplified correction errors due to improper reference location selection. At the same time, the SNR ranking results also provide a quantitative basis for the priority division of each acquisition location to be corrected, making the logical relationship between the subsequent correction order and signal quality more reasonable.

[0033] S02: Based on the spatial layout coordinates of the acoustic sensor, the preprocessed acoustic signal corresponding to the reference position, and the preprocessed acoustic signal corresponding to the acquisition position to be corrected, a propagation path difference matrix is ​​constructed. Based on the propagation path difference matrix, spatial feature correction is performed on the preprocessed acoustic signal corresponding to the acquisition position to be corrected. In this embodiment of the invention, the step of constructing a propagation path difference matrix based on the spatial layout coordinates of the voiceprint sensor, the preprocessed voiceprint signal corresponding to the reference position, and the preprocessed voiceprint signal corresponding to the acquisition position to be corrected, and performing spatial feature correction on the preprocessed voiceprint signal corresponding to the acquisition position to be corrected based on the propagation path difference matrix, includes: The spatial coordinates of the reference position and the spatial coordinates of each acquisition position to be calibrated are extracted from the spatial layout coordinates of each acoustic sensor. The Euclidean distance formula is used to calculate the spatial coordinates of the reference position and the spatial coordinates of each acquisition position to be calibrated, respectively, to obtain the spatial distance parameters of each acquisition position to be calibrated relative to the reference position.

[0034] Discrete Fourier transform is performed on the preprocessed acoustic signature signal sequence corresponding to the reference position and the preprocessed acoustic signature signal sequence corresponding to each acquisition position to be calibrated, respectively, to convert each signal sequence from time domain representation to frequency domain representation, thereby obtaining the frequency domain complex value sequence corresponding to each signal sequence. Amplitude information is extracted from each frequency domain complex value sequence to obtain the reference frequency domain amplitude spectrum corresponding to the reference position and the frequency domain amplitude spectrum to be calibrated corresponding to each acquisition position to be calibrated. At the same time, the original time domain waveform of each signal sequence is retained as the reference time domain waveform and the time domain waveform to be calibrated.

[0035] The time-domain waveform to be calibrated at each sampling location to be calibrated is cross-correlated with the reference time-domain waveform. That is, the reference time-domain waveform is used as a reference template, and the time-domain waveform to be calibrated is gradually slid along the time axis. The inner product value of the time-domain waveform to be calibrated and the reference time-domain waveform is calculated once for each sampling point slid. The sliding displacement corresponding to the maximum inner product value is recorded, and the sliding displacement value is determined as the arrival time difference of each sampling location to be calibrated relative to the reference location.

[0036] The amplitude ratio of the amplitude spectrum of the frequency domain to be calibrated to the reference frequency domain amplitude spectrum at each sampling location to be calibrated is calculated point by point. That is, at each discrete frequency point, the amplitude value of the amplitude spectrum of the frequency domain to be calibrated at that frequency point is divided by the amplitude value of the reference frequency domain amplitude spectrum at the same frequency point to obtain the amplitude ratio corresponding to each frequency point. The amplitude ratios of all frequency points are arranged in order from low to high frequency to form the relative amplitude difference coefficient of each sampling location to be calibrated relative to the reference location.

[0037] According to the spatial layout order of each acquisition location to be calibrated on the target cable, the spatial distance parameters corresponding to each acquisition location to be calibrated are arranged in the first column, the arrival time difference corresponding to each acquisition location to be calibrated is arranged in the second column, and the relative amplitude difference coefficient corresponding to each acquisition location to be calibrated is arranged in the third column. The above three columns of data are combined into a matrix data structure. Each row of the matrix corresponds to all the difference parameters of an acquisition location to be calibrated. The matrix as a whole is used as the propagation path difference matrix.

[0038] Extract the arrival time difference corresponding to each acquisition location to be calibrated from the propagation path difference matrix, and determine the sign of each arrival time difference in turn. When the arrival time difference is positive, determine that the time delay compensation direction corresponding to the acquisition location to be calibrated is forward. When the arrival time difference is negative, determine that the time delay compensation direction corresponding to the acquisition location to be calibrated is backward. When the arrival time difference is zero, determine that the acquisition location to be calibrated does not need to be compensated for time delay.

[0039] According to the determined time delay compensation direction, the preprocessed voiceprint signal sequence corresponding to each acquisition position to be corrected is shifted along the time axis. When the compensation direction is forward, the entire preprocessed voiceprint signal sequence is shifted towards the starting direction of the time axis by the number of sampling points equal to the absolute value of the arrival time difference. When the compensation direction is backward, the entire preprocessed voiceprint signal sequence is shifted towards the ending direction of the time axis by the number of sampling points equal to the absolute value of the arrival time difference, so that the voiceprint signal sequence of each acquisition position to be corrected is aligned with the voiceprint signal sequence of the reference position in time.

[0040] Extract the relative amplitude difference coefficients corresponding to each acquisition position to be corrected from the propagation path difference matrix, perform the reciprocal operation on the amplitude ratio of each frequency point in the relative amplitude difference coefficient, and use the reciprocal of the amplitude ratio of each frequency point as the amplitude compensation gain corresponding to each frequency point.

[0041] According to the calculated amplitude compensation gain, the amplitude value of the preprocessed acoustic signature signal at each frequency point corresponding to each acquisition location to be corrected is adjusted. That is, the amplitude value at each frequency point is multiplied by the amplitude compensation gain corresponding to that frequency point, so that the amplitude value of the preprocessed acoustic signature signal at each frequency point of each acquisition location to be corrected is consistent with the amplitude value of the preprocessed acoustic signature signal at each frequency point corresponding to the reference position. After the spatial feature correction is completed, the signal sequence corresponding to each acquisition location to be corrected is the corrected acoustic signature signal sequence.

[0042] The beneficial effect is that by systematically fusing the spatial layout coordinates of the acoustic fingerprint sensor with the time-frequency characteristics of the signal, a propagation path difference matrix is ​​constructed, which includes spatial distance parameters, time difference of arrival, and relative amplitude difference coefficients. Based on this matrix, time delay compensation and frequency domain amplitude compensation are performed on the signal to be corrected, realizing active correction of signal distortion caused by differences in propagation distance and medium. Compared with the existing technology that processes multiple sensor signals as independent data sources, the spatial feature correction of this invention enables the acoustic fingerprint signals at each acquisition location to achieve spatial consistency in the time and frequency domains. This eliminates the systematic deviation introduced by different propagation paths from the signal source, providing a high-quality input signal with time-frequency domain alignment for subsequent feature extraction and weighted fusion, and effectively suppressing the interference of propagation path differences on fault feature identification.

[0043] S03: Perform feature extraction on the preprocessed voiceprint signal and the corrected voiceprint signal corresponding to the reference position respectively to obtain the reference voiceprint feature parameter set and the corrected voiceprint feature parameter set; In this embodiment of the invention, the step of extracting features from the preprocessed voiceprint signal and the corrected voiceprint signal corresponding to the reference position to obtain a reference voiceprint feature parameter set and a corrected voiceprint feature parameter set includes: The Mel frequency cepstral coefficient analysis method is used to extract features from the preprocessed and corrected voiceprint signals corresponding to the reference position, respectively.

[0044] The number of sampling points in each frame (frame length) and the sampling point interval between the start points of adjacent frames (frame shift step) are pre-defined. The frame shift step is smaller than the frame length to ensure partial temporal overlap between adjacent frames. The preprocessed voiceprint signal corresponding to the reference position and each corrected voiceprint signal are processed into frames. A window with the frame length as its width is used to extract sampling points from the signal's starting point as the first frame. The window is then slid backward by one frame shift step to extract the second frame. This sliding and extraction operation is repeated until the window reaches the signal's endpoint, resulting in the voiceprint frame sequence corresponding to each signal.

[0045] The pre-emphasis processing is performed on each frame signal in the voiceprint frame sequence. Each frame signal is processed by a first-order high-pass filter. That is, for each sampling point in each frame signal, the amplitude of the sampling point is subtracted from the product of the amplitude of the previous sampling point and the pre-emphasis coefficient to obtain the pre-emphasis processed amplitude corresponding to the sampling point. After all sampling points are processed, the pre-emphasis signal corresponding to each frame is obtained.

[0046] A window function is applied to the pre-emphasis signal, and the window function sequence is multiplied with the pre-emphasis signal sequence according to the sampling point position, so that the amplitude at both ends of each frame signal gradually becomes zero, eliminating the signal discontinuity at the frame boundary, and obtaining the windowed signal corresponding to each frame.

[0047] Perform a Fast Fourier Transform on the windowed signal corresponding to each frame to convert the signal from the time domain to the frequency domain. Extract the amplitude part of the transformation result to obtain the frequency domain amplitude spectrum corresponding to each frame signal.

[0048] The amplitude of each frequency point in the frequency domain amplitude spectrum corresponding to each frame of signal is squared to obtain the power value of each frequency point. The power values ​​of all frequency points constitute the power spectral density estimate of each frame of signal.

[0049] The power spectral density estimate corresponding to each frame signal is input into a Mel filter bank composed of multiple triangular bandpass filters. The triangular bandpass filters are equally spaced on the Mel frequency scale. The power spectral density estimate is weighted and accumulated with the frequency response of each filter within the passband of each filter to obtain the energy value of each frame signal in each Mel frequency band, i.e., the frequency band energy value.

[0050] Take the natural logarithm of the energy values ​​of each Mel frequency band corresponding to each frame of signal, and convert each frequency band energy value into a logarithmic domain value to obtain the logarithmic energy spectrum data corresponding to each frame of signal.

[0051] A discrete cosine transform is performed on the logarithmic energy spectrum data corresponding to each frame of signal to convert each logarithmic energy spectrum from frequency domain representation to cepstral domain representation, resulting in a cepstral domain coefficient sequence. Low-order coefficients are selected from this coefficient sequence to form the Mel frequency cepstral coefficient eigenvector corresponding to each frame of signal.

[0052] Arrange the Mel frequency cepstral coefficient feature vectors of all voiceprint frames corresponding to the reference position in chronological order to obtain the reference voiceprint feature parameter set corresponding to the reference position. Arrange the Mel frequency cepstral coefficient feature vectors of all voiceprint frames of each corrected voiceprint signal in chronological order to obtain the corrected voiceprint feature parameter set corresponding to each corrected voiceprint signal.

[0053] The beneficial effect is that the Mel frequency cepstral coefficient analysis method is used to extract features from the preprocessed and corrected acoustic fingerprint signals corresponding to the reference position. Through processing steps such as framing, pre-emphasis, windowing, fast Fourier transform, Mel filtering, logarithmic operation, and discrete cosine transform, the original acoustic fingerprint signal is converted into a low-dimensional Mel frequency cepstral coefficient feature vector. This feature extraction method simulates the characteristics of human auditory perception, effectively extracting acoustic fingerprint feature components that are strongly correlated with the cable fault type, while suppressing redundant information unrelated to the fault. This results in the extracted reference acoustic fingerprint feature parameter set and the corrected acoustic fingerprint feature parameter set having good fault characterization ability and noise robustness, providing highly discriminative feature data for subsequent weighted fusion and fault matching.

[0054] After obtaining the frequency band energy value, the method further includes a step of calculating the correction consistency index based on the frequency band energy value, including: For each acquisition location to be corrected, the energy values ​​of each Mel frequency band corresponding to the acquisition location to be corrected and the energy values ​​of each Mel frequency band corresponding to the reference location are substituted into the cosine similarity calculation formula. The energy values ​​of each frequency band are used as the values ​​of each dimension of the vector to form the frequency band energy distribution vector. The cosine value of the angle between the frequency band energy distribution vector of the corrected voiceprint signal and the frequency band energy distribution vector of the preprocessed voiceprint signal corresponding to the reference location is calculated.

[0055] The formula for calculating the corrected consistency index is as follows: ; in, The number of filters in a Mel filter bank. For the first The frequency band energy value of the corrected acoustic signature signal at the passband output of each filter. For the first The frequency band energy value of the preprocessed acoustic signature signal corresponding to the reference position of the passband output of each filter. The corrected consistency index is denoted as .

[0056] The calibration consistency index ranges from 0 to 1. A value closer to 1 indicates a more consistent frequency band energy distribution between the calibrated acoustic signature signal and the reference location signal, resulting in a better calibration effect. Conversely, a value closer to 0 indicates a greater deviation between the two, leading to a worse calibration effect. The calibration consistency index serves as the basis for determining the dynamic fusion weights for each acquisition location to be calibrated in subsequent steps.

[0057] The beneficial effect is that the correction consistency index is obtained by calculating the cosine similarity between the frequency band energy distribution of the corrected acoustic signature signal and the reference position signal. This index quantitatively reflects the processing effect of the spatial feature correction stage. After normalizing the correction consistency index into dynamic fusion weights, channels with better correction effects receive a greater contribution in the fusion, while the contributions of channels with poorer correction effects are effectively suppressed, achieving adaptive optimization of the fusion strategy. Compared with the existing technology that determines fusion weights based on fixed empirical values ​​or simple signal quality indicators, the dynamic fusion weights of this invention are directly related to the actual correction effect of each channel, making the final fused acoustic signature feature parameter set participating in fault matching have a higher signal-to-noise ratio and fault feature representativeness, effectively improving the quality of the fused features.

[0058] As another optional implementation of this embodiment, after obtaining the frequency band energy values, a dynamic time warping algorithm can be used to calculate the similarity distance between the corrected voiceprint signal and the preprocessed voiceprint signal corresponding to the reference position in the frequency band energy sequence, and the reciprocal of this similarity distance is used as the correction consistency index. Specifically, the Mel frequency band energy values ​​of the corrected voiceprint signal are arranged in frequency band order to form a first energy sequence, and the Mel frequency band energy values ​​of the preprocessed voiceprint signal corresponding to the reference position are arranged in frequency band order to form a second energy sequence. The first energy sequence and the second energy sequence are used as two input sequences of the dynamic time warping algorithm, and the minimum cumulative distance that aligns the two sequences is calculated. The reciprocal of this minimum cumulative distance is used as the correction consistency index. This method is suitable for application scenarios where there is a local offset in the frequency band energy distribution.

[0059] S04: Perform weighted fusion of the reference voiceprint feature parameter set and the corrected voiceprint feature parameter set to obtain a fused voiceprint feature parameter set; In this embodiment of the invention, the step of weightedly fusing the reference voiceprint feature parameter set and the corrected voiceprint feature parameter set to obtain a fused voiceprint feature parameter set includes: The consistency indexes of each acquisition location to be calibrated are summed to obtain the sum of the consistency indices.

[0060] The ratio of the correction consistency index corresponding to each acquisition location to be corrected to the sum of the correction consistency indices is used as the dynamic fusion weight corresponding to each acquisition location to be corrected.

[0061] Extract the corrected voiceprint feature vectors corresponding to each acquisition position to be corrected from the set of corrected voiceprint feature parameters, and multiply the dynamic fusion weights corresponding to each acquisition position to be corrected with the corresponding corrected voiceprint feature vectors to obtain the weighted corrected voiceprint feature vectors corresponding to each acquisition position to be corrected.

[0062] The reference voiceprint feature vector corresponding to the reference position is extracted from the reference voiceprint feature parameter set. The reference voiceprint feature vector is added to the weighted corrected voiceprint feature vector of each acquisition position to be corrected according to each dimension of the feature vector. The sum of the reference vector value and all weighted vector values ​​in each dimension constitutes the fusion value of that dimension. The fusion values ​​of all dimensions form the fused voiceprint feature parameter set.

[0063] S05: Match the fused voiceprint feature parameter set with the preset fault voiceprint feature template library, generate graded early warning information based on the matching result, and output the graded early warning information to the monitoring terminal.

[0064] In this embodiment of the invention, the step of matching the fused voiceprint feature parameter set with a preset fault voiceprint feature template library, generating graded early warning information based on the matching result, and outputting the graded early warning information to the monitoring terminal includes: A fault acoustic signature template library is pre-stored on a server or vehicle-mounted terminal. This library contains standard acoustic signature parameter templates corresponding to various cable fault types, including insulation faults, short-circuit faults, and mechanical impact faults. The standard acoustic signature parameter templates are constructed as follows: typical acoustic signature samples of the corresponding fault type are collected in a laboratory environment or at a known fault site. The Mel-frequency cepstral coefficient feature vectors of each sample are extracted using the same feature extraction method as in S03. The mean value of each dimension of the feature vectors of all samples for the same fault type is taken, and this mean vector is used as the standard acoustic signature parameter template for that fault type.

[0065] Calculate the Euclidean distance between the fused voiceprint feature parameter set and the standard voiceprint feature parameter template corresponding to each fault type in the preset fault voiceprint feature template library. That is, take the square root of the sum of the squares of the differences between the values ​​of each dimension of the fused voiceprint feature parameter set and the values ​​of each dimension of the standard template to obtain the feature distance corresponding to the fault type.

[0066] The fault type with the smallest characteristic distance is selected as the current fault type of the target cable.

[0067] The minimum feature distance is obtained, and a normalization mapping process is performed on the minimum feature distance to obtain the fault confidence. The normalization mapping process adopts a preset exponential decay mapping method, specifically: the minimum feature distance is used as the independent variable of an exponential function, and the minimum feature distance is mapped from the distance space to the confidence space through the exponential function. The mapped confidence value is negatively correlated with the minimum feature distance; the smaller the minimum feature distance, the higher the fault confidence, and the larger the minimum feature distance, the lower the fault confidence. The decay coefficient of the exponential function is preset based on empirical values.

[0068] The warning level is determined based on the fault confidence level. A first warning level is defined as a fault confidence level greater than or equal to a first preset threshold; a second warning level is defined as a fault confidence level greater than or equal to a second preset threshold but less than the first preset threshold; and a third warning level is defined as a fault confidence level less than the second preset threshold. The first warning level indicates a high degree of fault certainty, requiring immediate response; the second warning level indicates a moderate probability of fault, requiring attention; and the third warning level indicates a low probability of fault, serving only as a notification. The specific values ​​of the first and second preset thresholds can be set by those skilled in the art based on the actual application scenario and the requirements for false alarm and false negative rates.

[0069] The fault type, the fault confidence level, and the warning level are encapsulated into hierarchical warning information. The structure of the hierarchical warning information is a data packet containing a fault type encoding field, a confidence level value field, and a warning level encoding field.

[0070] The graded early warning information is output to the monitoring terminal, which is a display device installed in the operation and maintenance center or on a handheld terminal device. The graded early warning information is sent to the monitoring terminal through a wired network or wireless network, and the monitoring terminal displays the fault type, fault confidence level and early warning level to the operation and maintenance personnel.

[0071] The beneficial effects are as follows: by matching the fused acoustic signature feature parameter set with a preset fault acoustic signature feature template library, the current fault type of the target cable is selected based on the minimum feature distance as the judgment criterion. The minimum feature distance is then normalized and mapped to obtain the fault confidence level. Based on the fault confidence level, the warning level is determined, and the graded warning information is output in the form of structured data packets, realizing a complete closed loop from data acquisition to warning output. Compared with the existing technology that relies on deep learning models for classification and recognition, this invention does not require large-scale labeled training data and model training processes. It has significant deployment advantages in actual engineering scenarios where historical fault samples are lacking. At the same time, the graded warning mechanism enables maintenance personnel to quickly judge the severity of the fault based on the warning level and take differentiated response measures, improving maintenance efficiency and emergency response capabilities.

[0072] As a preferred embodiment of this example, the construction of the fault acoustic signature feature template library in the above steps can also employ a feature extraction and clustering method based on deep learning. Specifically, a large number of cable acoustic signature samples with known fault types are collected. For each sample, the Mel-frequency cepstral coefficient feature vector is extracted as a sample feature according to the method in S03. All sample features are input into a deep feature extraction network based on the ResNet architecture. This network is trained under supervision using the cross-entropy loss function during the training phase. The training data is a set of acoustic signature feature samples labeled with fault types. After training, the deep discriminative feature vectors of each sample are extracted from the output of the global average pooling layer of the ResNet network. The mean of all deep discriminative feature vectors of the same fault type is taken in each dimension, and the mean vector is used as the standard acoustic signature feature parameter template for that fault type and stored in the template library. By automatically learning more discriminative feature representations through deep learning, the discriminative ability and generalization performance of the template library can be further improved.

[0073] like Figure 2 The diagram shown is a functional block diagram of a cable fault acoustic signature monitoring and early warning system based on feature domain reconstruction provided in an embodiment of the present invention.

[0074] The cable fault acoustic signature monitoring and early warning system based on feature domain reconstruction described in this invention can be installed in electronic devices. Depending on the functions implemented, the cable fault acoustic signature monitoring and early warning system based on feature domain reconstruction may include a signal acquisition module, a spatial correction module, a feature extraction module, a weighted fusion module, and an early warning output module. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and is stored in the memory of the electronic device.

[0075] In this embodiment, the functions of each module / unit are as follows: The signal acquisition module is used to acquire acoustic fingerprint signals generated during the operation of the target cable, perform data preprocessing, and select the reference position and the acquisition position to be corrected of the target cable based on the signal-to-noise ratio of the preprocessed acoustic fingerprint signal. The spatial correction module is used to construct a propagation path difference matrix based on the spatial layout coordinates of the acoustic sensor, the preprocessed acoustic signal corresponding to the reference position, and the preprocessed acoustic signal corresponding to the acquisition position to be corrected, and to perform spatial feature correction on the preprocessed acoustic signal corresponding to the acquisition position to be corrected based on the propagation path difference matrix. The feature extraction module is used to extract features from the preprocessed voiceprint signal and the corrected voiceprint signal corresponding to the reference position, respectively, to obtain a reference voiceprint feature parameter set and a corrected voiceprint feature parameter set. The weighted fusion module is used to perform weighted fusion on the reference voiceprint feature parameter set and the corrected voiceprint feature parameter set to obtain a fused voiceprint feature parameter set. The early warning output module is used to match the fused voiceprint feature parameter set with a preset fault voiceprint feature template library, generate graded early warning information based on the matching result, and output the graded early warning information to the monitoring terminal.

[0076] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0077] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0078] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for monitoring and early warning of cable fault acoustic signatures based on feature domain reconstruction, characterized in that, The method includes: S01: Collect the acoustic fingerprint signal generated during the operation of the target cable, perform data preprocessing, and select the reference position and the collection position to be calibrated of the target cable based on the signal-to-noise ratio of the preprocessed acoustic fingerprint signal. S02: Based on the spatial layout coordinates of the acoustic sensor, the preprocessed acoustic signal corresponding to the reference position, and the preprocessed acoustic signal corresponding to the acquisition position to be corrected, a propagation path difference matrix is ​​constructed. Based on the propagation path difference matrix, spatial feature correction is performed on the preprocessed acoustic signal corresponding to the acquisition position to be corrected. S03: Perform feature extraction on the preprocessed voiceprint signal and the corrected voiceprint signal corresponding to the reference position respectively to obtain the reference voiceprint feature parameter set and the corrected voiceprint feature parameter set; S04: Perform weighted fusion of the reference voiceprint feature parameter set and the corrected voiceprint feature parameter set to obtain a fused voiceprint feature parameter set; S05: Match the fused voiceprint feature parameter set with the preset fault voiceprint feature template library, generate graded early warning information based on the matching result, and output the graded early warning information to the monitoring terminal.

2. The cable fault acoustic signature monitoring and early warning method based on feature domain reconstruction as described in claim 1, characterized in that, The process involves acquiring acoustic signature signals generated during the operation of the target cable, performing data preprocessing, and selecting the reference position and the acquisition position to be calibrated for the target cable based on the signal-to-noise ratio of the preprocessed acoustic signature signals. Based on pre-deployed acoustic sensors, acoustic signals generated during the operation of the target cable are collected; The voiceprint signal is amplified, and the amplified analog voiceprint signal is converted from analog to digital to obtain a preprocessed voiceprint signal. Noise power estimation and signal power estimation are performed on the preprocessed acoustic signal respectively, and the signal-to-noise ratio corresponding to each acquisition location is calculated. The acquisition position corresponding to the highest signal-to-noise ratio was selected as the reference position; The remaining acquisition positions, excluding the reference position, are arranged in descending order of signal-to-noise ratio and marked as acquisition positions to be calibrated.

3. The cable fault acoustic signature monitoring and early warning method based on feature domain reconstruction as described in claim 1, characterized in that, The step of constructing a propagation path difference matrix based on the spatial layout coordinates of the voiceprint sensor, the preprocessed voiceprint signal corresponding to the reference position, and the preprocessed voiceprint signal corresponding to the acquisition position to be calibrated, and then performing spatial feature correction on the preprocessed voiceprint signal corresponding to the acquisition position to be calibrated based on the propagation path difference matrix, includes: The spatial distance between the reference position and the acquisition position to be corrected is measured to obtain the spatial distance parameter between the acquisition position to be corrected and the reference position. Time-frequency transformation is performed on the preprocessed acoustic fingerprint signal corresponding to the reference position and the preprocessed acoustic fingerprint signal corresponding to the acquisition position to be corrected, respectively. Extract the reference time-domain waveform and reference frequency-domain amplitude spectrum, as well as the time-domain waveform to be calibrated and the frequency-domain amplitude spectrum to be corrected from the time-frequency transformation results; Perform cross-correlation calculation on the time-domain waveform to be calibrated and the reference time-domain waveform to obtain the arrival time difference of the acquisition position to be calibrated relative to the reference position; The amplitude ratio between the amplitude spectrum of the frequency domain to be corrected and the amplitude spectrum of the reference frequency domain is calculated to obtain the relative amplitude difference coefficient between the acquisition position to be corrected and the reference position. Based on the spatial layout order of the acquisition locations to be corrected, the spatial distance parameters, the arrival time difference, and the relative amplitude difference coefficient are arranged in a matrix.

4. The cable fault acoustic signature monitoring and early warning method based on feature domain reconstruction as described in claim 3, characterized in that, The spatial feature correction of the preprocessed acoustic signature signal corresponding to the acquisition location to be corrected, based on the propagation path difference matrix, includes: Extract the arrival time difference corresponding to the acquisition position to be corrected from the propagation path difference matrix, and determine the time delay compensation direction corresponding to the acquisition position to be corrected based on the sign of the arrival time difference. Based on the time delay compensation direction, the preprocessed voiceprint signal corresponding to the acquisition position to be corrected is time-aligned. Extract the relative amplitude difference coefficients corresponding to the acquisition positions to be corrected from the propagation path difference matrix, and perform a reciprocal operation on the relative amplitude difference coefficients to obtain the amplitude compensation gain; Based on the amplitude compensation gain, the amplitude of the preprocessed acoustic signal corresponding to the acquisition position to be corrected is adjusted so that the amplitude value of the preprocessed acoustic signal at each frequency point is consistent with the preprocessed acoustic signal corresponding to the reference position.

5. The cable fault acoustic signature monitoring and early warning method based on feature domain reconstruction as described in claim 1, characterized in that, The step involves extracting features from the preprocessed and corrected voiceprint signals corresponding to the reference positions, respectively, to obtain a reference voiceprint feature parameter set and a corrected voiceprint feature parameter set, including: The Mel frequency cepstral coefficient analysis method was used to extract features from the preprocessed and corrected voiceprint signals corresponding to the reference positions, respectively. Based on the preset number of sampling points and the preset frame shift step size, the preprocessed voiceprint signal and the corrected voiceprint signal corresponding to the reference position are processed into frames to obtain a voiceprint frame sequence. The signal of each frame in the voiceprint frame sequence is pre-emphasized to compensate for the high-frequency energy loss of the signal during propagation, and a pre-emphasized signal is obtained. A window function is applied to the pre-emphasized signal to obtain a windowed signal; Perform a Fast Fourier Transform on the windowed signal to obtain the frequency domain amplitude spectrum; The power spectral density estimate is obtained by squaring the frequency domain amplitude spectrum.

6. The cable fault acoustic signature monitoring and early warning method based on feature domain reconstruction as described in claim 5, characterized in that, After obtaining the power spectral density estimate, the method further includes: The power spectral density estimate is input into the Mel filter bank to obtain the bandgap energy value; Taking the natural logarithm of the frequency band energy values ​​yields logarithmic energy spectrum data; Perform a discrete cosine transform on the logarithmic energy spectrum data, extract the low-order discrete cosine transform coefficients, and obtain the Mel frequency cepstral coefficient eigenvector. The Mel frequency cepstral coefficient feature vectors are aggregated in chronological order to obtain a reference voiceprint feature parameter set and a corrected voiceprint feature parameter set.

7. The cable fault acoustic signature monitoring and early warning method based on feature domain reconstruction as described in claim 6, characterized in that, After obtaining the frequency band energy value, the process also includes: Based on the frequency band energy value, the cosine similarity between the frequency band energy distribution of the corrected voiceprint signal and the frequency band energy distribution of the preprocessed voiceprint signal corresponding to the reference position is calculated to obtain the correction consistency index. The formula for calculating the corrected consistency index is as follows: ； in, The number of filters in a Mel filter bank. For the first The frequency band energy value of the corrected acoustic signature signal at the passband output of each filter. For the first The frequency band energy value of the preprocessed acoustic signature signal corresponding to the reference position of the passband output of each filter. The corrected consistency index is denoted as .

8. The cable fault acoustic signature monitoring and early warning method based on feature domain reconstruction as described in claim 7, characterized in that, The step of weightedly fusing the reference voiceprint feature parameter set and the corrected voiceprint feature parameter set to obtain a fused voiceprint feature parameter set includes: Calculate the calibration consistency index of the data acquisition locations to be calibrated; The ratio of the correction consistency index corresponding to the acquisition location to be corrected to the sum of the correction consistency indices is used as the dynamic fusion weight. Extract the corrected voiceprint feature vector from the set of corrected voiceprint feature parameters, and multiply the dynamic fusion weight with the corrected voiceprint feature vector to obtain the weighted corrected voiceprint feature vector. The reference voiceprint feature vector is extracted from the reference voiceprint feature parameter set, and the reference voiceprint feature vector and the weighted and corrected voiceprint feature vector are summed to obtain the fused voiceprint feature parameter set.

9. The cable fault acoustic signature monitoring and early warning method based on feature domain reconstruction as described in claim 1, characterized in that, The step of matching the fused voiceprint feature parameter set with a preset fault voiceprint feature template library, generating graded early warning information based on the matching result, and outputting the graded early warning information to the monitoring terminal includes: The feature distance between the fused voiceprint feature parameter set and the standard voiceprint feature parameter templates corresponding to each fault type in the preset fault voiceprint feature template library is calculated. The fault type with the smallest feature distance is selected as the current fault type of the target cable, and the minimum feature distance is normalized and mapped to obtain the fault confidence. Based on the fault confidence level, the warning level is determined, the fault type, the fault confidence level, and the warning level are encapsulated into graded warning information, and the graded warning information is output to the monitoring terminal.

10. A cable fault acoustic signature monitoring and early warning system based on feature domain reconstruction, characterized in that, The system for implementing the cable fault acoustic signature monitoring and early warning method based on feature domain reconstruction as described in claim 1 includes: The signal acquisition module is used to acquire acoustic fingerprint signals generated during the operation of the target cable, perform data preprocessing, and select the reference position and the acquisition position to be calibrated of the target cable based on the signal-to-noise ratio of the preprocessed acoustic fingerprint signal. The spatial correction module is used to construct a propagation path difference matrix based on the spatial layout coordinates of the acoustic sensor, the preprocessed acoustic signal corresponding to the reference position, and the preprocessed acoustic signal corresponding to the acquisition position to be corrected, and to perform spatial feature correction on the preprocessed acoustic signal corresponding to the acquisition position to be corrected based on the propagation path difference matrix. The feature extraction module is used to extract features from the preprocessed voiceprint signal and the corrected voiceprint signal corresponding to the reference position, respectively, to obtain a reference voiceprint feature parameter set and a corrected voiceprint feature parameter set; The weighted fusion module is used to perform weighted fusion on the reference voiceprint feature parameter set and the corrected voiceprint feature parameter set to obtain a fused voiceprint feature parameter set. The early warning output module is used to match the fused voiceprint feature parameter set with the preset fault voiceprint feature template library, generate graded early warning information based on the matching result, and output the graded early warning information to the monitoring terminal.

Images