Ultra-long distance cable fault monitoring method and system

Abstract

The invention discloses an ultra-long distance cable fault monitoring method and system, and belongs to the field of power system safety monitoring, and the method comprises the steps: obtaining a partial discharge high-frequency transient signal, a hyperspectral remote sensing image and atmospheric parameter data of a cable, carrying out the phase-space reconstruction of the partial discharge high-frequency transient signal, and constructing a chaotic feature; and performing atmospheric correction on the hyperspectral image based on the atmospheric parameters, generating a surface reflectance image, performing spectral unmixing on the surface reflectance image, and extracting a spectral fingerprint. The method comprises the following steps: performing space-time alignment on a partial discharge high-frequency transient signal and a hyperspectral remote sensing image to generate a space-time mapping index, and according to the space-time mapping index, pairing a chaotic feature with a spectral fingerprint to form a space-time feature pair; and predicting the fault probability of each position point of the cable in a future time range through the cross-modal fusion model, and judging a latent fault according to a threshold value. According to the invention, the problem of inaccurate monitoring caused by weak signals and index lag in the prior art can be solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system safety monitoring, in particular to a long-distance cable fault monitoring method and system. BACKGROUND

[0002] As the key artery of modern energy and information transmission, the fault monitoring of long-distance cable is the core task to ensure system safety. The existing monitoring of long-distance cable mainly relies on analyzing terminal electrical parameters to determine faults. However, this method is difficult to capture early fault precursors, and the fundamental reason is that the initial signal generated by early faults is extremely weak, and in long-distance transmission, it will be attenuated and dispersed, resulting in energy dissipation and waveform distortion, and finally completely submerged in system background noise. Traditional signal analysis tools are basically ineffective for such low signal-to-noise ratio nonlinear transient characteristics, resulting in a serious lack of early warning capability.

[0003] To improve the monitoring dimension, some technical solutions begin to try to fuse the internal electrical signals with the temperature data measured by the optical fiber laid along the cable body. However, this method requires additional physical sensing lines beside the cable, which is a huge and costly reconstruction project for the vast stock of cables. More importantly, temperature is a lagging indicator of faults - only when insulation deterioration develops to a certain extent and produces significant thermal effects can it be detected, and at this time the best preventive maintenance window has often been missed. Therefore, the existing fusion solution still has obvious technical bottlenecks in achieving real early and accurate prediction. SUMMARY

[0004] The embodiment of the present application provides a long-distance cable fault monitoring method and system, which can solve the problem of inaccurate monitoring caused by weak signals and lagging indicators in the prior art.

[0005] An embodiment of the present application provides a long-distance cable fault monitoring method, comprising: obtaining a partial discharge high-frequency transient signal of a cable to be monitored, a hyperspectral remote sensing image covering a cable line path, and atmospheric parameter data; reconstructing the phase space of the partial discharge high-frequency transient signal to generate a phase space attractor, and constructing a chaotic feature vector according to the maximum Lyapunov exponent and the fractal dimension of the phase space attractor; atmospheric correction is performed on the hyperspectral remote sensing image according to the atmospheric parameter data to generate a ground reflectance image; spectral unmixing is performed on the ground reflectance image according to a preset characteristic spectral model to extract a spectral fingerprint feature sequence; wherein the characteristic spectral model is a digital data model obtained by spectral measurement of microscopic discharge of a controlled cable sample; spatial mapping index is generated, and according to the spatial mapping index, a time-space feature pair sequence is generated by pairing the chaotic feature vector and the spectral fingerprint feature sequence; The time-space feature pair sequence is input into a preset cross-modal fusion model to generate a probability sequence of each position point of the cable to be monitored generating a fault in a future time range, and the probability sequences of all position points are combined to construct a time-space fault probability cloud map. It is judged whether the fault probability of any spatial position point in the time-space fault probability cloud map is continuously higher than a preset threshold in a future time period; if yes, it is determined that the cable to be monitored has a latent fault, and a fault warning is triggered; if not, it is determined that the cable to be monitored does not have a latent fault.

[0006] Further, the phase space reconstruction is performed on the partial discharge high-frequency transient signal to generate a phase space attractor, including: The partial discharge high-frequency transient signal is sampled at a fixed time interval to generate a sampling sequence; The sampling sequence is de-trended to obtain a de-trended signal sequence; The de-trended signal sequence is amplitude normalized to obtain a preprocessed signal sequence; Based on the preprocessed signal sequence, a self-correlation function curve is calculated to obtain self-correlation function data; The first minimum point of the self-correlation function data is determined, and the time corresponding to the first minimum point is determined as the reconstruction delay time; Based on the preprocessed signal sequence, the best embedding dimension is determined by the false neighbor point method; According to the reconstruction delay time and the best embedding dimension, the preprocessed signal sequence is delay sampled to generate a multi-dimensional state vector sequence; The multi-dimensional state vector sequence is arranged in time sequence to form a phase space trajectory, and the phase space trajectory is taken as a phase space attractor.

[0007] Further, the atmospheric parameter data is used to correct the hyperspectral remote sensing image to generate a ground reflectance image, including: The original pixel value of the hyperspectral remote sensing image is radiometrically calibrated to generate radiometric brightness data; Based on the atmospheric parameter data, the air scattering influence on the radiometric brightness data is calculated according to a preset atmospheric radiation transmission model to generate scattering correction data; Based on the atmospheric parameter data, the atmospheric absorption influence on the radiometric brightness data is calculated according to a preset atmospheric radiation transmission model to generate absorption correction data; correcting the radiation intensity data according to the scattering correction data and the absorption correction data, to generate an atmosphere-corrected surface reflectance image.

[0008] Further, the surface reflectance image is spectrally un-mixed according to a preset characteristic spectral model to extract a spectral fingerprint feature sequence, including: extracting light intensity values of each spectral band from the preset characteristic spectral model to construct a target spectral vector; traversing each pixel point along the cable path to be monitored in the surface reflectance image and extracting reflectance values of each pixel point in each spectral band to generate a corresponding pixel spectral vector; for each pixel spectral vector, a matching filter algorithm is used to calculate a projection value of the current pixel spectral vector in the direction of the target spectral vector to generate a projection result of each pixel; normalizing the projection results to generate corresponding abundance values; arranging the abundance values in the order of the cable path to form a spectral fingerprint feature sequence.

[0009] Further, the partial discharge high-frequency transient signal, the hyperspectral remote sensing image, and the preset GIS path map of the cable to be monitored are spatio-temporally aligned to generate a spatio-temporal mapping index, including: detecting events of the partial discharge high-frequency transient signal to determine occurrence times of each partial discharge event to generate an event time list; for each partial discharge event, determining an occurrence position of the event on the cable to be monitored based on the partial discharge high-frequency transient signal to generate an event spatial position list; according to the event time list and the event spatial position list, pairing the time and the corresponding spatial position of each partial discharge event to generate an event spatio-temporal pair list; mapping the spatial position in the event spatio-temporal pair list to a corresponding hyperspectral image pixel position through the preset GIS path map of the cable to be monitored to generate an event pixel correspondence table; combining the event spatio-temporal pair list and the event pixel correspondence table to generate a spatio-temporal mapping index.

[0010] Further, the chaotic feature vector and the spectral fingerprint feature sequence are paired according to the spatio-temporal mapping index to generate a spatio-temporal feature pair sequence, including: traversing each record in the spatio-temporal mapping index to obtain the occurrence time and the corresponding pixel position of each event from the record; According to the occurrence time of the current event, a chaotic feature vector corresponding to the current event is extracted from the chaotic feature vector time sequence; according to the pixel position of the current event, a spectral fingerprint feature value corresponding to the current event is extracted from the spectral fingerprint feature sequence; The chaotic feature vector and the spectral fingerprint feature value are paired to form a space-time feature pair for each event; According to the spatial order of the cable path to be monitored, all space-time feature pairs are arranged to generate a space-time feature pair sequence.

[0011] Further, the cross-modal fusion model includes a chaotic feature encoding module, a spectral feature encoding module, an attention fusion module, and a fault probability prediction module; The space-time feature pair sequence is input into a preset cross-modal fusion model to generate a probability sequence of faults generated by each position point of the cable to be monitored within a future time range, including: The space-time feature pair sequence is input into a preset cross-modal fusion model to enable the cross-modal fusion model to perform deep feature encoding on the chaotic feature vectors in the space-time feature pair sequence through the chaotic feature encoding module to generate a chaotic modal deep feature sequence; The spectral fingerprint feature sequence in the space-time feature pair sequence is deep feature encoded through the spectral feature encoding module to generate a spectral modal deep feature sequence; The chaotic modal deep feature sequence and the spectral modal deep feature sequence are weighted and fused through the attention fusion module to generate a fused space-time feature vector sequence; The fused space-time feature vector sequence is used for time series prediction through the fault probability prediction module to generate a probability sequence of faults generated by each position point of the cable to be monitored within a future time range.

[0012] Further, the training of the cross-modal fusion model includes: Obtain a cable fault historical data set; wherein the cable fault historical data set includes a plurality of training data, each training data including a historical space-time feature pair sequence and a real fault state label in a future time period corresponding to the historical space-time feature pair sequence in time; According to a preset batch size, the cable fault historical data set is randomly divided into a plurality of batches of training samples; The training samples in batches are sequentially input into the cross-modal fusion model, and end-to-end iterative training is performed on all learnable parameters in the cross-modal fusion model until a preset number of training rounds is reached; wherein, when receiving a batch of training samples, the cross-modal fusion model encodes the chaotic feature vector of the historical spatio-temporal feature pair sequence in the current batch of training samples through the chaotic feature encoding module to generate a chaotic modal deep feature sequence in the training process; encodes the spectral fingerprint feature sequence of the historical spatio-temporal feature pair sequence in the current batch of training samples through the spectral feature encoding module to generate a spectral modal deep feature sequence in the training process; The attention fusion module performs weighted fusion according to the chaotic modal deep feature sequence in the training process and the spectral modal deep feature sequence in the training process to generate a fused spatio-temporal feature vector sequence in the training process; The fault probability prediction module performs time series prediction according to the fused spatio-temporal feature vector sequence in the training process to generate a prediction probability sequence corresponding to the current training sample for generating a fault in a future time range; a loss function value is calculated according to the prediction probability sequence and the corresponding real fault state label through a preset loss function; the learnable network parameters in the chaotic feature encoding module, the spectral feature encoding module, the attention fusion module and the fault probability prediction module are updated through gradient backpropagation and update using a preset optimizer.

[0013] Further, the feature spectrum model is constructed in the following way: Obtain laboratory spectrum data; wherein, the laboratory spectrum data is obtained by applying an excitation to a cable sample of the same model as the cable to be monitored in a controlled experimental device to induce local micro-discharge, and using a spectrometer to collect optical radiation signals of the discharge phenomenon; Noise filtering and baseline correction are performed on the laboratory spectrum data to generate standardized spectrum sample data; In the standardized spectrum sample data, a peak detection algorithm is used to identify and extract feature spectral lines representing local micro-discharge phenomena; The wavelength and corresponding relative light intensity values of the feature spectral lines are stored in data to construct a feature spectrum model.

[0014] On the basis of the above method embodiment, the present application provides a system embodiment.

[0015] An embodiment of the present application provides a kind of super-long distance cable fault monitoring system, comprising: data acquisition module, chaos feature extraction module, spectral feature extraction module, feature pairing module, cross-modal fusion module and fault prediction module; The data acquisition module is used to acquire the partial discharge high-frequency transient signal of the cable to be monitored, the hyperspectral remote sensing image covering the cable line path and the atmospheric parameter data; The chaos feature extraction module is used to reconstruct the phase space of the partial discharge high-frequency transient signal, generate a phase space attractor, and construct a chaos feature vector according to the maximum Lyapunov exponent and the fractal dimension of the phase space attractor. The spectral feature extraction module is used to perform atmospheric correction on the hyperspectral remote sensing image according to the atmospheric parameter data to generate a ground reflectance image, and perform spectral unmixing on the ground reflectance image according to a preset feature spectral model to extract a spectral fingerprint feature sequence. The feature spectral model is a digital data model obtained by spectral measurement of microscopic discharge of a controlled cable sample. The feature pairing module is used to perform time-space alignment on the partial discharge high-frequency transient signal, the hyperspectral remote sensing image and a preset GIS path map of the cable to be monitored to generate a time-space mapping index, and pair the chaos feature vector with the spectral fingerprint feature sequence according to the time-space mapping index to generate a time-space feature pair sequence. The cross-modal fusion module is used to input the time-space feature pair sequence into a preset cross-modal fusion model to generate a probability sequence of fault generation at each position point of the cable to be monitored within a future time range, and combine the probability sequences of all position points to construct a time-space fault probability cloud map. The fault prediction module is used to determine whether the fault probability of any spatial position point in the time-space fault probability cloud map is continuously higher than a preset threshold value in a future time period. If yes, it is determined that the cable to be monitored has a latent fault, and a fault warning is triggered. If no, it is determined that the cable to be monitored does not have a latent fault.

[0016] Compared with the prior art, the present application has the following beneficial effects: The embodiment of the present application provides a kind of super long distance cable fault monitoring method and system.The method obtains the partial discharge high-frequency transient signal, hyperspectral remote sensing image and atmospheric parameter data of cable to be monitored;Partial discharge signal is reconstructed in phase space, and chaotic characteristic vector is constructed;Hyperspectral image is corrected in atmosphere, and according to the characteristic spectral model constructed according to the characteristic spectral line obtained from laboratory micro-discharge, spectral fingerprint feature sequence is extracted;Chaotic characteristic vector and spectral fingerprint feature sequence are spatio-temporally aligned in combination with cable GIS path, and spatio-temporal feature pair sequence is generated;Spatio-temporal feature pair sequence is input into cross-modal fusion model, the future failure probability of each position point is predicted, and spatio-temporal failure probability cloud chart is constructed;According to the cloud chart, it is judged whether the failure probability of any spatial point is continuously more than threshold value, so as to determine latent fault and trigger early warning.

[0017] The signal generated by early cable fault is usually weak and difficult to capture, and the essential reason is that micro-discharge is the initial physical process, which changes the local electric field and signal fluctuation, but does not cause significant changes in cable temperature or structure, so it is difficult to find potential failure in advance by traditional monitoring based on macroscopic indicators. In view of this feature, the present application starts from the micro level, analyzes the internal signal of the cable based on chaos theory, quantitatively describes the dynamic characteristics of the signal evolution from stationary to chaos, and forms the early warning of internal signal dynamics. At the same time, by using hyperspectral remote sensing image combined with micro-discharge spectral model obtained in laboratory, local optical signals generated by micro-discharge are extracted by spectral unmixing and matching filter algorithm. These optical signals reflect the micro-discharge process itself, not the cable structure damage, and can expose potential failure earlier than temperature indicators that rely on heat accumulation. Finally, internal chaotic characteristics and external spectral fingerprints are comprehensively analyzed by spatio-temporal alignment and cross-modal fusion, realizing early failure warning of double micro-precursors, and improving the sensitivity and accuracy of monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a flowchart of a super long distance cable fault monitoring method provided by an embodiment of the present application.

[0019] Figure 2 is a structural schematic diagram of a super long distance cable fault monitoring system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0021] It should be noted that the present application aims to solve the core problem in the existing ultra-long distance cable fault monitoring technology, i.e. the extremely weak early fault precursor signal, and the traditional monitoring means mostly rely on macroscopic and lagging physical indicators, resulting in delayed early warning and insufficient reliability. To fundamentally break through this bottleneck, the core technical concept of the present application is to sink the monitoring focus from macroscopic fault phenomena to microscopic physical precursors, and establish a dual early warning system based on the mutual verification of internal electrical dynamics evolution and external physical environment products.

[0022] In this system, for the analysis of internal electrical signals, the present application realizes that the most initial form of fault is microscopic discharge, which will break the original stable and linear rules of the signal. Therefore, the present application introduces chaos theory as a nonlinear analysis tool, and quantitatively describes the dynamic process of the signal evolution from order to chaos state through methods such as phase space reconstruction, so as to capture the earliest signs of system instability that cannot be identified by traditional methods.

[0023] In addition, to obtain independent external physical evidence, the present application innovatively applies hyperspectral remote sensing technology to microscopic precursor detection. Instead of observing the macroscopic damage of the cable body, it aims to identify and extract the spectral fingerprints of specific chemical substances generated by the ionization of the surrounding air in the moment of microscopic discharge. The fingerprint is a direct and instantaneous physical product of the fault, and has incomparable early and sensitivity compared to lagging indicators such as temperature which requires heat accumulation.

[0024] The final technical effect of the present application is achieved by deeply fusing the above two independent microscopic precursor characteristics. By spatiotemporal alignment, the internal chaotic characteristics and the external spectral fingerprints are accurately paired, and a cross-modal fusion model is used for comprehensive analysis, which can realize the cross verification of the dual evidence. This fundamentally solves the problem of false positives or false negatives caused by single source monitoring, greatly improves the accuracy and reliability of fault monitoring while greatly advancing the early warning window.

[0025] As shown in Figure 1 To solve the problem of inaccurate monitoring caused by weak signals and lagging indicators in the prior art, an embodiment of the present application provides an ultra-long distance cable fault monitoring method, which at least includes the following steps: Step S1, obtaining the partial discharge high-frequency transient signal of the cable to be monitored, the hyperspectral remote sensing image covering the cable line path, and the atmospheric parameter data; Specifically, in one specific embodiment of the present application, the initial step of the ultra-long distance cable fault monitoring method is the synchronous acquisition of multi-source heterogeneous data, which provides a data basis for subsequent fusion analysis.

[0026] Specifically, the monitoring system continuously collects the partial discharge high-frequency transient signals inside the cable through high-frequency current transformers or capacitive coupling sensors deployed at the cable line terminal or key nodes along the line. The signals are the direct current or voltage response generated in the electrical circuit when micro-discharge occurs in the insulating medium inside the cable. The transient waveform and timing characteristics contain nonlinear dynamic information that can represent the evolution process of the cable insulation state from stable to deterioration, which is the fundamental basis for subsequent chaos feature analysis.

[0027] At the same time, the system calls or controls the airborne or satellite remote sensing platform equipped with a hyperspectral imager to obtain hyperspectral remote sensing images that completely cover the geographic path of the cable to be monitored within a preset monitoring period. The hyperspectral remote sensing images record the reflection spectrum information of the ground surface along the cable in hundreds of continuous, narrow wavebands on the sun's radiation. This information provides a core optical data source for accurately identifying and extracting specific physical and chemical signs, i.e., spectral fingerprints, generated by micro-discharge activities in the external environment of the cable.

[0028] To ensure the accuracy of spectral fingerprint feature extraction, atmospheric parameter data covering the cable line path also need to be obtained synchronously. The atmospheric parameter data includes key information such as ambient temperature, humidity, visibility, and atmospheric aerosol optical depth corresponding to the imaging time of the hyperspectral remote sensing images. These data can be obtained from the ground weather station network along the line or the public weather service data interface. The atmospheric parameter data is the core input for subsequent atmospheric radiation transfer model calculation and atmospheric correction. Its role is to accurately strip the scattering and absorption effects of the atmosphere on the ground reflectance, which is a key prerequisite for restoring the true ground reflectance and ensuring the effectiveness of subsequent spectral unmixing.

[0029] The three data sources obtained through the above steps form a multi-dimensional heterogeneous data set composed of internal electrical signals, external optical images, and environmental correction parameters, which provides a comprehensive and reliable data foundation for subsequent deep fusion warning through dual micro-precursors.

[0030] Step S2, reconstructing the phase space of the partial discharge high-frequency transient signal to generate a phase space attractor, and constructing a chaos feature vector according to the maximum Lyapunov exponent and fractal dimension of the phase space attractor; In a preferred embodiment, the phase space reconstruction of the partial discharge high-frequency transient signal to generate a phase space attractor includes: sampling the partial discharge high-frequency transient signal at a fixed time interval to generate a sample sequence; detrending the sample sequence to obtain a detrended signal sequence; amplitude normalization is performed on the detrended signal sequence to obtain a preprocessed signal sequence; Based on the preprocessed signal sequence, the autocorrelation function curve is calculated to obtain the autocorrelation function data; The first minimum point of the autocorrelation function data is determined, and the time corresponding to the first minimum point is determined as the reconstruction delay time; Based on the preprocessed signal sequence, the best embedding dimension is determined by using the false neighbor point method; According to the reconstruction delay time and the best embedding dimension, the preprocessed signal sequence is sampled with delay to generate a multi-dimensional state vector sequence; The multi-dimensional state vector sequence is arranged in time sequence to form a phase space trajectory, and the phase space trajectory is taken as a phase space attractor.

[0031] Specifically, after obtaining the partial discharge high-frequency transient signal representing the cable state, a core step of the embodiment is to use chaos theory to perform in-depth dynamic analysis on the signal. Although the original one-dimensional time sequence signal contains fault information, the internal and complex dynamic law is difficult to directly reveal. Therefore, the one-dimensional signal is expanded to a high-dimensional state space by the phase space reconstruction technology, so as to restore and visualize the internal structure of the original dynamic system, that is, the phase space attractor.

[0032] In a preferred embodiment, the specific process of the above phase space reconstruction first samples the obtained continuous partial discharge high-frequency transient signal at a fixed sampling frequency, and converts it into a discrete time sequence signal, that is, a sampling sequence. In order to eliminate the potential influence of low-frequency interference such as direct current bias or power frequency trend on nonlinear analysis, the sampling sequence needs to be detrended to obtain a zero-mean detrended signal sequence. Subsequently, in order to facilitate subsequent mathematical calculation and eliminate the influence of signal amplitude change, amplitude normalization is performed on the detrended signal sequence to obtain a preprocessed signal sequence with a uniform amplitude interval.

[0033] After the signal preprocessing is completed, two key parameters of phase space reconstruction need to be determined: reconstruction delay time and best embedding dimension . In the embodiment, the delay time is determined by the autocorrelation function method. First, the autocorrelation function curve of the preprocessed signal sequence is calculated; the point at which the function value on the curve first decreases to a minimum value can well ensure that the linear correlation of the reconstructed dimensions is the lowest, so the time corresponding to the minimum value is determined as the reconstruction delay time . The best embedding dimension It is determined by the false neighbor method. The principle of this method is that when the embedding dimension is too low, the projection of the original dynamic system attractor in the low-dimensional space will produce false neighbor points. By calculating and observing the change of the false neighbor rate as the dimension increases, the dimension at which the false neighbor rate first drops to a preset low threshold (usually close to zero) is selected as the minimum integer dimension that can fully unfold the true structure of the attractor, that is, the optimal embedding dimension. .

[0034] The reconstruction delay time determined by the above method and the optimal embedding dimension , the preprocessed signal sequence can be reconstructed. This reconstruction process is based on Takens' embedding theorem, which constructs a multidimensional state vector sequence by delayed sampling. Its mathematical expression is as follows:

[0035] Where, is the first state vector; is the first sampling points; is the optimal embedding dimension; Reconstruction delay time.

[0036] All calculated The multidimensional state vector sequence is arranged in time order, and the trajectory formed is the phase space trajectory, which fully presents the attractor form of the original dynamic system.

[0037] After the phase space attractor is successfully generated, this embodiment further calculates its quantitative chaos characteristic index.

[0038] To quantitatively assess the predictability of a system, this embodiment calculates the maximum Lyapunov exponent of a phase space attractor. This exponent quantifies the average exponential rate at which two infinitely adjacent points on the attractor's trajectory separate or converge over time. A positive maximum Lyapunov exponent is a clear indicator of chaotic behavior. This embodiment employs the Small Data Method for calculations. The core concept is to track the average logarithmic distance growth rate of pairs of adjacent points on the attractor. The basic principle can be expressed as follows:

[0039] Where, is the estimated value of the maximum Lyapunov exponent; is the time evolution step; is the total number of evolution steps; is the first distance between adjacent points after evolution time . is the first distance between adjacent points. If the calculated , it indicates that the system trajectory has local instability and is in a chaotic state. The larger the value is, the worse the predictability of the system is, and the higher the degree of chaos of the signal is.

[0040] To quantitatively evaluate the complexity of the system, the fractal dimension of the phase space attractor is calculated in this embodiment, and specifically, the correlation dimension can be used as an effective estimate thereof. The correlation dimension is calculated by the G-P algorithm (Grassberger-Procaccia Algorithm), and the core is to calculate the probability that the distance between any two points on the attractor is less than a given scale , that is, the correlation integral . For an attractor with fractal characteristics, the correlation integral and the distance scale have a power law relationship:

[0041] In the formula, is the correlation integral, indicating the probability that the distance between two points in the phase space is less than . is the distance scale; is the correlation dimension.

[0042] In practice, by calculating the slope of the linear part (i.e., the scaling region) of the curve of and in the logarithmic coordinate system, the correlation dimension can be obtained. The value of the correlation dimension reflects the complexity of the geometric structure of the attractor and the number of independent variables contained in the signal. A non-integer, finite correlation dimension value is also strong evidence that the system has chaotic characteristics.

[0043] Finally, the maximum Lyapunov exponent, which is a scalar eigenvalue, and the correlation dimension, which is also a scalar eigenvalue, are combined to form a two-dimensional chaotic feature vector.

[0044] Through the above series of processes, the original, high-dimensional and complex transient signal is successfully converted into a low-dimensional chaotic feature vector that can accurately quantify the stability of the system, providing a highly sensitive and quantitative first warning feature for the subsequent cross-modal fusion of the internal state of the cable.

[0045] Step S3, performing atmospheric correction on the hyperspectral remote sensing image according to the atmospheric parameter data to generate a surface reflectance image; performing spectral unmixing on the surface reflectance image according to a preset characteristic spectral model to extract a spectral fingerprint feature sequence; wherein the characteristic spectral model is a digital data model obtained by performing spectral measurement on micro-discharge of a controlled cable sample; In a preferred embodiment, the atmospheric correction on the hyperspectral remote sensing image according to the atmospheric parameter data to generate a surface reflectance image comprises: performing radiometric calibration on original pixel values of the hyperspectral remote sensing image to generate radiance data; based on the atmospheric parameter data, calculating air scattering influence on the radiance data according to a preset atmospheric radiation transfer model to generate scattering correction data; based on the atmospheric parameter data, calculating atmospheric absorption influence on the radiance data according to a preset atmospheric radiation transfer model to generate absorption correction data; performing correction on the radiance data according to the scattering correction data and the absorption correction data to generate a surface reflectance image after atmospheric correction.

[0046] In a preferred embodiment, the spectral unmixing on the surface reflectance image according to a preset characteristic spectral model to extract a spectral fingerprint feature sequence comprises: extracting light intensity numerical values of each spectral band from the preset characteristic spectral model to construct a target spectral vector; traversing each pixel point along a cable path to be monitored in the surface reflectance image and extracting reflectance values of each pixel point in each spectral band to generate a corresponding pixel spectral vector; for each pixel spectral vector, using a matched filter algorithm to calculate a projection value of the current pixel spectral vector in the direction of the target spectral vector to generate a projection result of each pixel; normalizing each projection result to generate a corresponding abundance value; arranging each abundance value in the order of the cable path to form a spectral fingerprint feature sequence.

[0047] In a preferred embodiment, the characteristic spectral model is constructed by: obtaining laboratory spectral data; wherein the laboratory spectral data is obtained by applying excitation to a cable sample of the same model as the cable to be monitored in a controlled experimental device to induce local micro-discharge, and using a spectrometer to collect optical radiation signals of the discharge phenomenon; performing noise filtering and baseline correction on the laboratory spectral data to generate standardized spectral sample data; In the standardized spectral sample data, a characteristic spectral line representing a local micro-discharge phenomenon is identified and extracted by a peak detection algorithm; The wavelength of the characteristic spectral line and the corresponding relative light intensity value are stored in data form to construct a characteristic spectral model.

[0048] Specifically, after obtaining the hyperspectral remote sensing image and the synchronous atmospheric parameter data, another key technical route of the present application is to extract the external physical evidence capable of representing the micro-discharge phenomenon, i.e., the spectral fingerprint feature sequence. The core of this process is to restore and identify the weak spectral features caused by micro-discharge, which are the real concern, from the mixed light signals received by the sensor in space, containing complex information of the atmosphere and the earth's surface, through a series of refined processing.

[0049] The processing process first needs to perform atmospheric correction to eliminate the interference of the atmosphere on the optical signal and generate the real earth's surface reflectance image. In a preferred embodiment, the atmospheric correction process includes: first, performing radiometric calibration on the original pixel value (DN value) of the hyperspectral remote sensing image to convert it into radiance data at the entrance pupil of the sensor, which has physical meaning. However, the radiance data contains the whole process information of the solar radiation passing through the atmosphere to the earth's surface, being reflected by the earth's surface and then passing through the atmosphere to be received by the sensor, which is seriously affected by the scattering and absorption of air molecules. Therefore, the preset atmospheric radiation transfer model (such as MODTRAN, 6S, etc.) must be used to correct the radiance data together with the synchronous acquisition of atmospheric parameter data (including temperature, humidity, aerosol optical depth, etc.). The physical model of the correction process can be simplified as:

[0050] In the formula, is the earth's surface reflectance to be solved; is the total radiance value received by the sensor after radiometric calibration; is the atmospheric path radiance, i.e., the part directly entering the sensor by atmospheric scattering itself; is the total atmospheric transmittance from the earth's surface to the sensor; is the total solar irradiance reaching the earth's surface.

[0051] In this embodiment, can be directly obtained from the image after radiometric calibration. The atmospheric path radiance , the total atmospheric transmittance and the total solar irradiance can be accurately calculated by the atmospheric radiation transfer model with the synchronous acquisition of the atmospheric parameter data as input. Through the above calculation, the atmospheric influence can be removed to obtain the real earth's surface reflectance image.

[0052] After obtaining the ground reflectance image, the next step is to extract the spectral fingerprint sequence from the pre-set characteristic spectral model. In a preferred embodiment, the extraction process employs a matched filter algorithm. The algorithm is an ideal detector for detecting a known target signal from a mixed signal background. First, a target spectral vector representing the micro-discharge phenomenon is extracted from the pre-set characteristic spectral model. Then, for each pixel along the cable path in the ground reflectance image, a pixel spectral vector is extracted by collecting its reflectance values in all spectral bands. The core of the matched filter algorithm is to maximize the response of the target signal while suppressing the response of the background noise, and the output result can be calculated by the following formula:

[0053] wherein, is the output result of the matched filter; is the current pixel spectral vector to be detected; is the target spectral vector extracted from the characteristic spectral model; is the mean spectral vector of the background pixels; is the inverse of the spectral covariance matrix of the background pixels.

[0054] In the present embodiment, the pixel spectral vector is obtained from the ground reflectance image, the target spectral vector is obtained from the pre-set characteristic spectral model, and the background statistics and are obtained by statistical calculation on the background region of the image. The projection result is normalized to form the spectral fingerprint sequence.

[0055] The pre-set characteristic spectral model is constructed by one or more processors performing the following steps: first, a set of laboratory spectral data is obtained; wherein the laboratory spectral data is obtained by applying an excitation to a cable sample of the same model as the cable to be monitored in a controlled experimental device to induce local micro-discharge, and using a spectrometer to collect the optical radiation signal of the discharge phenomenon. Then, the laboratory spectral data is digitally processed, including noise filtering and baseline correction, to generate standardized spectral sample data. Next, in the standardized spectral sample data, one or more characteristic spectral lines that can uniquely represent the local micro-discharge phenomenon are identified and extracted by means of a peak detection algorithm or the like. Finally, the wavelengths of the characteristic spectral lines and their corresponding relative intensity values are stored digitally to construct a pre-set characteristic spectral model containing the data characteristic spectral lines.

[0056] In summary, by a series of refined data processing such as atmospheric correction, characteristic spectral model construction and matching filtering, the original hyperspectral remote sensing image seriously interfered by the atmosphere is successfully converted into a spectral fingerprint feature sequence which can sensitively reflect the microscopic discharge precursor and accurately corresponds to the spatial position of the cable line, thereby providing a high-quality and high-sensitivity second warning feature for subsequent cross-modal fusion.

[0057] In step S4, the partial discharge high-frequency transient signal, the hyperspectral remote sensing image and the preset GIS path map of the cable to be monitored are spatiotemporally aligned to generate a spatiotemporal mapping index, and the chaotic feature vector and the spectral fingerprint feature sequence are paired according to the spatiotemporal mapping index to generate a spatiotemporal feature pair sequence. In one preferred embodiment, the spatiotemporal alignment of the partial discharge high-frequency transient signal, the hyperspectral remote sensing image and the preset GIS path map of the cable to be monitored to generate the spatiotemporal mapping index comprises: Event detection is performed on the partial discharge high-frequency transient signal to determine the occurrence time of each partial discharge event, and an event time list is generated. For each partial discharge event, the occurrence position of the event on the cable to be monitored is determined based on the partial discharge high-frequency transient signal, and an event spatial position list is generated. According to the event time list and the event spatial position list, the time and the corresponding spatial position of each partial discharge event are paired to generate an event spatiotemporal pair list. The spatial positions in the event spatiotemporal pair list are mapped to the corresponding hyperspectral image pixel positions through the preset GIS path map of the cable to be monitored to generate an event pixel correspondence table. The event spatiotemporal pair list and the event pixel correspondence table are combined to generate the spatiotemporal mapping index.

[0058] In one preferred embodiment, the pairing of the chaotic feature vector and the spectral fingerprint feature sequence according to the spatiotemporal mapping index to generate the spatiotemporal feature pair sequence comprises: Each record in the spatiotemporal mapping index is traversed to obtain the occurrence time and the corresponding pixel position of each event from the record. For each event, the chaotic feature vector corresponding to the current event is extracted from the chaotic feature vector time sequence according to the occurrence time of the current event, and the spectral fingerprint feature value corresponding to the current event is extracted from the spectral fingerprint feature sequence according to the pixel position of the current event. The chaotic feature vector and the spectral fingerprint feature value are paired to form the spatiotemporal feature pair of each event. All spatiotemporal feature pairs are arranged in the spatial order of the cable path to be monitored to generate the spatiotemporal feature pair sequence.

[0059] Specifically, after the chaotic feature vector representing the internal state of the cable and the spectral fingerprint sequence representing the external environmental physical precursors are extracted respectively, a core step of the present application is to accurately align and pair the two heterogeneous, time and domain separated data streams. The fundamental purpose is to establish a unique and determined correspondence between each internal electrical disturbance event (described by the chaotic feature vector) and the physical precursors (described by the spectral fingerprint feature value) that may be generated at the same space-time position outside the cable. This process is the key prerequisite for subsequent effective cross-modal fusion and dual evidence cross-validation.

[0060] In a preferred embodiment, the space-time alignment process is realized by constructing a space-time mapping index. First, event detection needs to be performed on the continuously collected partial discharge high-frequency transient signals. This detection process can identify and separate each independent transient pulse event caused by microscopic discharge from the background noise through digital signal processing methods such as short-time energy analysis, wavelet transform, or setting dynamic threshold, thereby determining the exact occurrence time of each partial discharge event and forming an event time list.

[0061] Next, each detected partial discharge event needs to be spatially located. When sensors are evenly distributed at both ends of the cable terminal, the Time Difference of Arrival (TDOA) method can be used to determine the occurrence position of the event on the cable line. This method is based on the physical principle that electromagnetic waves propagate at a nearly constant speed in the cable medium. The positioning calculation can be represented by the following formula:

[0062] wherein, is the physical distance from the fault event to the cable terminal; is the total length of the cable segment measured from both ends; is the propagation speed of the transient signal in the cable; is the time difference value of the same discharge event signal arriving at the cable terminal and terminal.

[0063] In this embodiment, the cable segment length and the signal propagation speed are known or pre-calibrated parameters, and the time difference value The direct measurement can be obtained from the signals collected synchronously at both ends. Through the calculation, the physical mileage position of each discharge event on the line can be determined, forming a list of spatial positions of events. Then, the list of event times is matched with the list of spatial positions of events to form a list of event space-time pairs. Finally, each physical mileage position in the list is converted into a geographical spatial coordinate through a preset GIS path map of the cable to be monitored, which has been geographically matched, and is further mapped to the corresponding hyperspectral image pixel position, to ultimately generate a space-time mapping index containing information such as event time, physical position and pixel position.

[0064] After the space-time mapping index is generated, the separated chaotic feature vectors and the spectral fingerprint feature sequence can be matched according to the index. In a preferred embodiment, the matching process includes: traversing each record in the space-time mapping index to obtain the occurrence timestamp and the corresponding pixel position of each local discharge event. For each record in the index, the system finds and extracts the chaotic feature vector corresponding to the timestamp from the chaotic feature vector time sequence generated previously and arranged along the time axis. At the same time, according to the pixel position of the record, the spectral fingerprint feature value corresponding to the pixel position is found and extracted from the spectral fingerprint feature sequence generated previously and arranged along the spatial path. The chaotic feature vector and the spectral fingerprint feature value thus extracted are combined to form a “space-time feature pair” that can completely describe the double precursors inside and outside the discharge event. This process is repeated for all events recorded in the space-time mapping index, and all generated space-time feature pairs are arranged in the spatial order of the cable path to ultimately form a complete and structured space-time feature pair sequence.

[0065] Through the above series of steps of space-time alignment and feature matching, the present application successfully converts two independent original feature data streams organized in different dimensions into a unified, highly structured space-time feature pair sequence, each data point of which contains double precursor information, laying a neat and reliable data foundation for subsequent input into a cross-modal fusion model for high-precision prediction analysis.

[0066] Step S5, inputting the space-time feature pair sequence into a preset cross-modal fusion model to generate a probability sequence of each position point of the cable to be monitored generating a fault in a future time range, and combining the probability sequences of all position points to construct a space-time fault probability cloud map; In a preferred embodiment, the cross-modal fusion model includes a chaotic feature encoding module, a spectral feature encoding module, an attention fusion module and a fault probability prediction module. inputting the spatiotemporal feature pair sequence into a preset cross-modal fusion model to generate a probability sequence of each position point of the cable to be monitored generating a fault within a future time range, comprising: inputting the spatiotemporal feature pair sequence into the preset cross-modal fusion model, so that the cross-modal fusion model encodes the chaotic feature vector in the spatiotemporal feature pair sequence through the chaotic feature encoding module to generate a chaotic modal deep feature sequence; encoding the spectral fingerprint feature sequence in the spatiotemporal feature pair sequence through the spectral feature encoding module to generate a spectral modal deep feature sequence; weighting and fusing the chaotic modal deep feature sequence and the spectral modal deep feature sequence through the attention fusion module to generate a fused spatiotemporal feature vector sequence; performing time series prediction on the fused spatiotemporal feature vector sequence through the fault probability prediction module to generate a probability sequence of each position point of the cable to be monitored generating a fault within a future time range.

[0067] In a preferred embodiment, the training of the cross-modal fusion model comprises: obtaining a cable fault historical data set; wherein the cable fault historical data set comprises a plurality of training data, each training data comprising a historical spatiotemporal feature pair sequence and a true fault state label within a future time period corresponding to the historical spatiotemporal feature pair sequence in time; randomly dividing the cable fault historical data set into a plurality of batches of training samples according to a preset batch size; inputting the batches of training samples into the cross-modal fusion model in sequence to perform end-to-end iterative training on all learnable parameters in the cross-modal fusion model until a preset number of training rounds is reached; wherein the cross-modal fusion model, upon receiving each batch of training samples, encodes the chaotic feature vector of the historical spatiotemporal feature pair sequence in the current batch of training samples through the chaotic feature encoding module to generate a chaotic modal deep feature sequence in the training process; encoding the spectral fingerprint feature sequence of the historical spatiotemporal feature pair sequence in the current batch of training samples through the spectral feature encoding module to generate a spectral modal deep feature sequence in the training process; weighting and fusing the chaotic modal deep feature sequence in the training process and the spectral modal deep feature sequence in the training process through the attention fusion module to generate a fused spatiotemporal feature vector sequence in the training process; The fault probability prediction module performs time series prediction according to the fused spatio-temporal feature vector sequence in the training process, and generates a prediction probability sequence corresponding to the current training sample, which indicates the possibility of generating a fault in a future time range; a preset loss function is used to calculate a loss function value according to the prediction probability sequence and a corresponding real fault state label; and a preset optimizer is used to perform gradient back propagation and update on the learnable network parameters in the chaotic feature encoding module, the spectral feature encoding module, the attention fusion module and the fault probability prediction module according to the loss function value.

[0068] Specifically, after generating the structured spatio-temporal feature pair sequence through spatio-temporal alignment and pairing, the core task of the present application is to use a preset cross-modal fusion model to perform deep analysis and time series prediction on the sequence. The fundamental purpose is to explore the complex and nonlinear internal relationship between the internal chaotic feature and the external spectral fingerprint, and based on this relationship, to make a quantitative and probabilistic accurate prediction of the fault risk of the cable in a future time range.

[0069] In a preferred embodiment, the cross-modal fusion model is a deep neural network model, the internal structure of which is carefully designed to adapt to the dual-source heterogeneous data characteristics of the present application. The model mainly includes a chaotic feature encoding module, a spectral feature encoding module, an attention fusion module and a fault probability prediction module. When receiving the spatio-temporal feature pair sequence, the inference (i.e. prediction) process of the model is as follows: first, the chaotic feature encoding module and the spectral feature encoding module, as two parallel “feature extractors”, process the input chaotic feature vector sequence and spectral fingerprint feature sequence respectively. These two sub-modules can adopt structures such as one-dimensional convolutional neural network (1D-CNN) or recurrent neural network (RNN), aiming to learn and extract higher-dimensional deep features that can represent the spatial correlation of the fault signs along the line from the respective input sequences, and generate chaotic modal deep feature sequences and spectral modal deep feature sequences respectively.

[0070] Subsequently, the attention fusion module dynamically weights and fuses the two deep feature sequences. The core role of this module is to enable the model to adaptively learn the relative importance of chaotic features and spectral features at each position point of the cable. For example, in some working conditions, the disturbance of the electrical signal may be more critical; while in other working conditions, the spectral anomaly of the external environment may provide more decisive evidence. This dynamic weighting process can be realized by a gating mechanism:

[0071]

[0072] In the formula, is the index of the location point along the cable path; is the deep feature vector of the chaotic modality at the location point ; is the deep feature vector of the spectral modality at the location point ; represents concatenating two feature vectors; is the learnable weight matrix in the gating network; is the learnable bias term in the gating network; is the Sigmoid activation function, which generates a weight gate ranging from 0 to 1 ; represents element-wise multiplication; is the spatio-temporal feature vector generated after attention-weighted fusion at the location point .

[0073] Combining the feature vectors of all location points, we obtain the fused spatio-temporal feature vector sequence. Finally, the fault probability prediction module (which can be composed of one or more fully connected layers) receives this fused feature sequence, performs nonlinear transformation and time series analysis, and finally outputs a probability sequence that represents the probability of each location point failing in the future time range. Visualizing this probability sequence along the cable path can construct an intuitive and dynamic spatio-temporal fault probability cloud map.

[0074] In a preferred embodiment, the above cross-modal fusion model is trained in a supervised learning manner. The training process includes: first, obtaining a cable fault historical data set composed of a large number of training samples. Each training sample includes a "historical spatio-temporal feature pair sequence" extracted from historical monitoring data within a known time period, and a "true fault status label" (e.g., 0 for no fault, 1 for fault) in a future time window after the time period. The data set is randomly divided into several batches of training samples according to the preset batch size.

[0075] During training, each batch of training samples is input into the model in turn for end-to-end iterative training. Upon receiving each batch of training samples, the model first performs a forward propagation step identical to the inference process described above, generating a "predicted probability sequence" of future fault status. Subsequently, a pre-set loss function is used to quantify the difference between the "predicted probability sequence" and the "true fault status label". For such binary classification probability prediction problems, a binary cross-entropy (Binary Cross-Entropy) loss function can be used:

[0076] wherein, is the calculated loss function value; is the total number of position points in the current batch; is the true fault status label (0 or 1) of the position point . is the predicted probability output by the model for the position point .

[0077] In the present embodiment, is obtained from the cable fault history dataset, is output in real time by the model during forward propagation. Finally, using a preset optimizer (such as the Adam optimizer), all four modules (chaotic feature encoding module, spectral feature encoding module, attention fusion module, and fault probability prediction module) in the model are updated synchronously and jointly according to the calculated loss function value by the gradient backpropagation algorithm. This iteration process is repeated until the prediction performance of the model converges on the validation set or the preset number of training rounds is met.

[0078] Through the construction, inference, and training of the deep learning model described above, the present application can automatically learn and establish a precise fault prediction model from complex, multi-modal microscopic precursor data, ultimately achieving high-sensitivity, high-reliability early warning of latent faults.

[0079] Step S6, determine whether the fault probability of any spatial position point in the spatiotemporal fault probability cloud map is continuously higher than the preset threshold in the future time period; if yes, it is determined that the cable to be monitored has a latent fault, and a fault warning is triggered; if not, it is determined that the cable to be monitored does not have a latent fault.

[0080] Specifically, after generating the spatiotemporal fault probability cloud map representing the future fault risk of each position point along the cable line by the cross-modal fusion model, the last step of the present application is to establish an objective, reliable, and automated fault diagnosis and warning decision mechanism based on the probability cloud map. The core purpose is to convert the continuous probability value output by the model into a clear diagnostic conclusion of "latent fault" or "no latent fault" that can be directly used by operation and maintenance personnel, and trigger the corresponding operation.

[0081] Simply judging according to the instantaneous fault probability value is easy to be affected by accidental noise or non-fault transient disturbance, and may cause false alarm of the early warning. In order to solve this problem, a preferred embodiment of the present application adopts a joint judgment criterion based on the "amplitude-time length" double conditions. The criterion aims to filter out random and short-term probability fluctuations and only alarms the abnormal point position showing persistence and high risk, so as to ensure the high reliability of the early warning. The criterion needs to preset two key thresholds: a fault probability threshold and a duration threshold . The two thresholds can be set by those skilled in the art according to the historical operation and maintenance data of the cable, the safe operation regulations and the trade-off between the early warning sensitivity and the false alarm rate.

[0082] The logic of the joint judgment criterion can be formally described by the following condition expression. For any spatial position point in the space-time fault probability cloud map, the early warning state at a future time is triggered if and only if the following conditions are met:

[0083] In the formula, is the fault probability prediction value of the spatial position point output by the cross-modal fusion model at a certain time in the past ; is the current monitoring time; is any historical time within a sliding time window with the current monitoring time as the end point; is a preset duration threshold for defining the length of the sliding time window; is a preset fault probability threshold.

[0084] In the present embodiment, the value of which is continuously output by the cross-modal fusion model. When the system monitors that, for any spatial position point , all the fault probability prediction values at all times within a duration of are higher than the preset probability threshold , the result of the above condition expression is true.

[0085] Once the early warning state If the error is judged as true, the system determines that the monitored cable has a latent fault at that location. The system then automatically triggers the fault warning mechanism. This warning mechanism may include generating a warning report containing key information such as the fault warning level, the precise spatial coordinates (or physical distance) of the suspected fault location, the current fault probability, and the chaotic characteristics and spectral fingerprints used to trigger the warning. This warning report is then sent to the operations and maintenance manager via the monitoring system interface, SMS text message, or a designated network protocol. If the warning status remains false for all spatial locations throughout the entire monitoring cycle, the system determines that the monitored cable has no latent faults and continues silent monitoring.

[0086] Through the above-mentioned joint judgment criteria based on the dual conditions of "amplitude-duration", the present invention constructs an intelligent "early warning filter", which ensures that only those fault precursors with a high degree of certainty and a continuous deterioration trend will be confirmed as latent faults. While ensuring extremely high monitoring sensitivity, it effectively suppresses the false alarm rate and provides highly reliable and operational early warning decision support for operation and maintenance personnel.

[0087] Based on the above method embodiments, the present invention provides corresponding system embodiments.

[0088] like Figure 2 As shown, an embodiment of the present invention provides an ultra-long distance cable fault monitoring system, comprising: a data acquisition module, a chaos feature extraction module, a spectral feature extraction module, a feature matching module, a cross-modal fusion module and a fault prediction module; The data acquisition module is used to obtain the high-frequency transient signal of partial discharge of the cable to be monitored, the hyperspectral remote sensing image covering the cable line path, and the atmospheric parameter data; The chaos feature extraction module is used to perform phase space reconstruction on the partial discharge high-frequency transient signal, generate a phase space attractor, and construct a chaos feature vector based on the maximum Lyapunov exponent and fractal dimension of the phase space attractor; The spectral feature extraction module is used to perform atmospheric correction on the hyperspectral remote sensing image based on atmospheric parameter data to generate a surface reflectance image; perform spectral unmixing on the surface reflectance image based on a preset characteristic spectrum model to extract a spectral fingerprint feature sequence; wherein the characteristic spectrum model is a digital data model obtained by spectrally measuring the microscopic discharge of the controlled cable sample; The feature pairing module is used to perform spatiotemporal alignment on the partial discharge high-frequency transient signal, the hyperspectral remote sensing image, and the preset GIS path map of the cable to be monitored to generate a spatiotemporal mapping index, and to pair the chaotic feature vector with the spectral fingerprint feature sequence based on the spatiotemporal mapping index to generate a spatiotemporal feature pair sequence; The cross-modal fusion module is configured to input the sequence of pairs of spatiotemporal features into a preset cross-modal fusion model to generate a probability sequence of each position point of the cable to be monitored generating a fault in a future time range, and combine the probability sequences of all position points to construct a spatiotemporal fault probability cloud map. The fault prediction module is configured to determine whether the fault probability of any spatial position point in the spatiotemporal fault probability cloud map is continuously higher than a preset threshold in a future time period, and if yes, determine that the cable to be monitored has a latent fault and trigger a fault warning, and if no, determine that the cable to be monitored does not have a latent fault.

[0089] It should be noted that the above-described embodiments of the system correspond to the above-described embodiments of the application, and can implement any of the above-described cable fault monitoring methods of the application. In addition, the above-described embodiments of the system are only illustrative, and the modules described as separate components can or can not be physically separated, and the components displayed as modules can or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. Some or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. In addition, in the system embodiment provided by the application, the connection relationship between the modules indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0090] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described can be combined in any suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine the different embodiments or examples described in the specification and the features of the different embodiments or examples without contradiction.

[0091] The above describes the preferred embodiments of the present application. It should be noted that those skilled in the art can make several improvements and refinements without departing from the principles of the present application, and these improvements and refinements are also considered within the scope of protection of the present application.

Claims

1. A method for monitoring ultra-long distance cable faults, characterized in that: include: Obtain high-frequency transient signals of partial discharge of the cable to be monitored, hyperspectral remote sensing images covering the cable line path, and atmospheric parameter data; The phase space of the high-frequency transient signal of partial discharge is reconstructed to generate the phase space attractor, and the chaotic eigenvector is constructed according to the maximum Lyapunov exponent and fractal dimension of the phase space attractor. Atmospheric correction is performed on hyperspectral remote sensing images based on atmospheric parameter data to generate surface reflectance images. Spectral unmixing is performed on the surface reflectance images based on a preset characteristic spectrum model to extract a spectral fingerprint feature sequence. The characteristic spectrum model is a digital data model obtained by spectrally measuring the microscopic discharge of a controlled cable sample. The high-frequency transient signals of partial discharge, hyperspectral remote sensing images, and the preset GIS route map of the cable to be monitored are aligned in time and space to generate a time-space mapping index. Based on the time-space mapping index, the chaotic feature vectors are paired with the spectral fingerprint feature sequence to generate a time-space feature pair sequence. The spatiotemporal feature pair sequence is input into a preset cross-modal fusion model to generate a probability sequence of faults occurring at each location point of the cable to be monitored within a future time range. The probability sequences of all location points are then combined to construct a spatiotemporal fault probability cloud map. Determine whether the failure probability of any spatial location point in the time-space fault probability cloud map is continuously higher than the preset threshold in the future time period; if so, it is determined that there is a latent fault in the monitored cable and a fault warning is triggered; if not, it is determined that there is no latent fault in the monitored cable.

2. The method for monitoring ultra-long distance cable faults according to claim 1, wherein: The phase space reconstruction of the partial discharge high-frequency transient signal to generate a phase space attractor includes: Sampling the partial discharge high-frequency transient signal at fixed time intervals to generate a sampling sequence; performing detrending processing on the sampling sequence to obtain a detrended signal sequence; performing amplitude normalization processing on the detrended signal sequence to obtain a preprocessed signal sequence; Calculating an autocorrelation function curve based on the preprocessed signal sequence to obtain autocorrelation function data; Determining a first minimum point of the autocorrelation function data, and determining a time corresponding to the first minimum point as a reconstruction delay time; Based on the preprocessed signal sequence, determining the optimal embedding dimension using a false neighbor method; performing delayed sampling on the preprocessed signal sequence according to the reconstruction delay time and optimal embedding dimension to generate a multidimensional state vector sequence; The multi-dimensional state vector sequence is arranged in time order to form a phase space trajectory, and the phase space trajectory is used as a phase space attractor.

3. The ultra-long distance cable fault monitoring method according to claim 1, characterized in that: The atmospheric correction of the hyperspectral remote sensing image according to the atmospheric parameter data to generate the surface reflectance image includes: Performing radiometric calibration on the original pixel values ​​of the hyperspectral remote sensing image to generate radiometric brightness data; Based on the atmospheric parameter data, and in accordance with a preset atmospheric radiation transmission model, calculating the air scattering effect on the radiance data to generate scattering correction data; Based on the atmospheric parameter data, and in accordance with a preset atmospheric radiation transmission model, calculating the atmospheric absorption effect on the radiance data to generate absorption correction data; The radiation brightness data is corrected according to the scattering correction data and the absorption correction data to generate a surface reflectance image that has been corrected for atmosphere.

4. The method for monitoring ultra-long distance cable faults according to claim 1, wherein: The method of performing spectral unmixing on the surface reflectance image according to a preset characteristic spectrum model and extracting a spectral fingerprint feature sequence includes: Extracting the light intensity value of each spectral band from the preset characteristic spectrum model to construct a target spectrum vector; Traversing each pixel point along the cable path to be monitored in the surface reflectance image, and extracting the reflectance value of each pixel point in each spectral band to generate a corresponding pixel spectral vector; For each pixel spectral vector, a matched filtering algorithm is used to calculate the projection value of the current pixel spectral vector in the direction of the target spectral vector, and generate the corresponding projection result for each pixel; Each projection result is normalized to generate the corresponding abundance value; The abundance values ​​are arranged in the order of the cable path to form a spectral fingerprint feature sequence.

5. The ultra-long distance cable fault monitoring method according to claim 1, wherein: The method of performing spatiotemporal alignment of the partial discharge high-frequency transient signal, the hyperspectral remote sensing image, and the preset GIS path map of the cable to be monitored to generate a spatiotemporal mapping index includes: Performing event detection on the partial discharge high-frequency transient signal, determining the occurrence time of each partial discharge event, and generating an event time list; For each partial discharge event, based on the partial discharge high-frequency transient signal, determine the occurrence location of the event on the cable to be monitored and generate a list of event spatial locations; According to the event time list and the event spatial position list, pairing the time of each partial discharge event with the corresponding spatial position to generate an event time-space pair list; Mapping the spatial positions in the event space-time pair list to the corresponding hyperspectral image pixel positions through a preset GIS path map of the cable to be monitored, and generating an event pixel correspondence table; The event space-time pair list is combined with the event pixel correspondence table to generate the space-time mapping index.

6. The method for monitoring ultra-long distance cable faults according to claim 1, wherein: The method of pairing the chaotic feature vector with the spectral fingerprint feature sequence according to the spatiotemporal mapping index to generate a spatiotemporal feature pair sequence includes: Traverse each record in the spatiotemporal mapping index and obtain the occurrence time and corresponding pixel position of each event from the record; For each event, according to the occurrence time of the current event, the chaotic feature vector corresponding to the current event is extracted from the chaotic feature vector time series; according to the pixel position of the current event, the spectral fingerprint feature value corresponding to the current event is extracted from the spectral fingerprint feature sequence; Pairing the chaotic feature vector with the spectral fingerprint feature value to form a spatiotemporal feature pair for each event; All the spatiotemporal feature pairs are arranged according to the spatial order of the cable paths to be monitored to generate a spatiotemporal feature pair sequence.

7. The method for monitoring ultra-long distance cable faults according to claim 1, wherein: The cross-modal fusion model includes a chaos feature encoding module, a spectral feature encoding module, an attention fusion module and a fault probability prediction module; The process of inputting the sequence of spatiotemporal feature pairs into a preset cross-modal fusion model to generate a probability sequence of faults occurring at each location point of the cable to be monitored within a future time range includes: Inputting the spatiotemporal feature pair sequence into a preset cross-modal fusion model, so that the cross-modal fusion model performs deep feature encoding on the chaotic feature vectors in the spatiotemporal feature pair sequence through the chaotic feature encoding module to generate a chaotic modal deep feature sequence; Performing deep feature encoding on the spectral fingerprint feature sequence in the spatiotemporal feature pair sequence through the spectral feature encoding module to generate a spectral modality deep feature sequence; By means of the attention fusion module, weighted fusion is performed according to the chaotic modal deep feature sequence and the spectral modal deep feature sequence to generate a fused spatiotemporal feature vector sequence; The fault probability prediction module performs time series prediction based on the fused spatiotemporal feature vector sequence to generate a probability sequence of faults occurring at each location point of the cable to be monitored within a future time range.

8. The method for monitoring ultra-long distance cable faults according to claim 7, wherein: The training of the cross-modal fusion model includes: Acquire a cable fault history dataset; wherein the cable fault history dataset includes a plurality of training data, each training data including a historical spatiotemporal feature pair sequence and a real fault state label in a future time period corresponding to the historical spatiotemporal feature pair sequence in time; According to a preset batch size, the cable fault history dataset is randomly divided into several batches of training samples; Inputting each batch of training samples into the cross-modal fusion model in sequence, and performing end-to-end iterative training on all learnable parameters in the cross-modal fusion model until a preset number of training rounds is reached; wherein, when the cross-modal fusion model receives each batch of training samples, the chaotic feature vector of the historical spatiotemporal feature pair sequence in the current batch of training samples is deep-feature encoded by the chaotic feature encoding module to generate a chaotic modal deep feature sequence during the training process; Through the spectral feature encoding module, deep feature encoding is performed on the spectral fingerprint feature sequence of the historical spatiotemporal feature pair sequence in the current batch of training samples to generate a spectral modality deep feature sequence during the training process; By means of the attention fusion module, weighted fusion is performed according to the chaotic modal deep feature sequence and the spectral modal deep feature sequence in the training process to generate a fused spatiotemporal feature vector sequence in the training process; Through the fault probability prediction module, time series prediction is performed based on the fused spatiotemporal feature vector sequence in the training process to generate a predicted probability sequence of faults generated within a future time range corresponding to the current training sample; through a preset loss function, the loss function value is calculated based on the predicted probability sequence and the corresponding true fault state label; using a preset optimizer, gradient backpropagation and update are performed on the learnable network parameters in the chaotic feature encoding module, the spectral feature encoding module, the attention fusion module and the fault probability prediction module according to the loss function value.

9. The method for monitoring ultra-long distance cable faults according to claim 1, wherein: The characteristic spectrum model is constructed in the following way: Acquiring laboratory spectral data; wherein the laboratory spectral data is obtained by applying excitation to a cable sample of the same model as the cable to be monitored in a controlled experimental device to induce local microscopic discharge, and using a spectrometer to collect optical radiation signals of the discharge phenomenon; performing noise filtering and baseline correction on the laboratory spectral data to generate standardized spectral sample data; In the standardized spectrum sample data, a peak detection algorithm is used to identify and extract characteristic spectrum lines representing local microscopic discharge phenomena; The wavelength of the characteristic spectrum line and the corresponding relative light intensity value are stored in digital form to construct a characteristic spectrum model.

10. An ultra-long distance cable fault monitoring system, characterized in that: include: Data acquisition module, chaos feature extraction module, spectral feature extraction module, feature pairing module, cross-modal fusion module and fault prediction module; The data acquisition module is used to obtain the high-frequency transient signal of partial discharge of the cable to be monitored, the hyperspectral remote sensing image covering the cable line path, and the atmospheric parameter data; The chaos feature extraction module is used to perform phase space reconstruction on the partial discharge high-frequency transient signal, generate a phase space attractor, and construct a chaos feature vector based on the maximum Lyapunov exponent and fractal dimension of the phase space attractor; The spectral feature extraction module is used to perform atmospheric correction on the hyperspectral remote sensing image based on atmospheric parameter data to generate a surface reflectance image; perform spectral unmixing on the surface reflectance image based on a preset characteristic spectrum model to extract a spectral fingerprint feature sequence; wherein the characteristic spectrum model is a digital data model obtained by spectrally measuring the microscopic discharge of the controlled cable sample; The feature pairing module is used to perform spatiotemporal alignment on the partial discharge high-frequency transient signal, the hyperspectral remote sensing image, and the preset GIS path map of the cable to be monitored to generate a spatiotemporal mapping index, and to pair the chaotic feature vector with the spectral fingerprint feature sequence based on the spatiotemporal mapping index to generate a spatiotemporal feature pair sequence; The cross-modal fusion module is used to input the sequence of spatiotemporal feature pairs into a preset cross-modal fusion model to generate a probability sequence of faults occurring at each location point of the cable to be monitored within a future time range, and to combine the probability sequences of all location points to construct a spatiotemporal fault probability cloud map; The fault prediction module is used to determine whether the failure probability of any spatial location point in the spatiotemporal fault probability cloud map is continuously higher than a preset threshold in the future time period; if so, it is determined that there is a latent fault in the cable to be monitored and a fault warning is triggered; if not, it is determined that there is no latent fault in the cable to be monitored.

Images