High-voltage cable multi-parameter intelligent diagnosis method, system, device and storage medium

Abstract

The application relates to the technical field of cable partial discharge detection and diagnosis, and provides a high-voltage cable multi-parameter intelligent diagnosis method, system, device and storage medium, so as to solve the problems of low accuracy and poor reliability of high-voltage cable multi-parameter diagnosis caused by multiple factors. The method comprises the following steps: decomposing and filtering a partial discharge signal after interference is removed, extracting a time-domain amplitude change rate and an energy distribution parameter, and generating a multi-dimensional time-frequency domain feature vector; calibrating the multi-dimensional time-frequency domain feature vector, performing nonlinear manifold mapping, generating a low-dimensional feature vector matched with dielectric spectrum characteristics, matching a partial discharge type with the highest similarity to a recognition result, combining a final loss threshold interval to determine an insulation deterioration grade, and generating a comprehensive diagnosis report. The application captures a partial discharge signal matched with a dielectric spectrum through a very high frequency sensor array, realizes accurate recognition of a discharge type and dynamic evaluation of an insulation deterioration grade through multi-scale analysis and deep learning fusion, and improves the reliability of high-voltage cable diagnosis.

Description

Technical Field

[0001] The present application relates to the technical field of cable partial discharge detection and diagnosis, and in particular to a multi-parameter intelligent diagnosis method and system for high-voltage cables. Background Art

[0002] High-voltage cables are key components of power systems. Partial discharge (PD) caused by insulation degradation is a key cause of cable failures. In complex electromagnetic environments and under strong power frequency interference, real-time online monitoring of the insulation condition at cable connectors and terminals is essential. This allows for dynamic identification of PD types and quantification of degradation, while ensuring signal robustness, robust characterization, and interpretable diagnostic results to support proactive grid operation and maintenance decisions.

[0003] The existing solution uses a fixed-frequency sensor to collect partial discharge signals, combines empirical mode decomposition to extract time-frequency domain features, performs feature dimensionality reduction through principal component analysis, and then inputs the signal into a support vector machine model for classification and identification. Finally, the insulation degradation level is determined based on the threshold method.

[0004] However, the existing scheme uses sensors with fixed frequency bands, which are prone to loss of effective signal components due to mismatch with the dielectric spectrum of cable insulation materials; empirical mode decomposition is easily affected by modal aliasing, feature extraction is not robust enough, and principal component analysis dimensionality reduction is difficult to characterize the nonlinear manifold structure of the discharge signal, resulting in low discrimination of similar discharge modes by the classification model, degradation assessment is easily interfered by feature redundancy, and the misjudgment rate is high. Summary of the Invention

[0005] The present application provides a high-voltage cable multi-parameter intelligent diagnosis method and system to solve the problem of low accuracy and poor reliability of high-voltage cable multi-parameter diagnosis caused by multiple factors in the prior art.

[0006] In a first aspect, the present application provides a multi-parameter intelligent diagnosis method for high-voltage cables, comprising:

[0007] Deploy corresponding ultra-high frequency sensor arrays at the joints and terminals of the high-voltage cable to obtain partial discharge signals within the insulation layer of the high-voltage cable. The operating frequency band of the ultra-high frequency sensor array matches the dielectric spectrum characteristics of the high-voltage cable insulation material.

[0008] The local discharge signal in the insulation layer of the high-voltage cable is filtered out of power frequency and low-frequency interference through a Butterworth high-pass filter, the local discharge signal after interference filtering is decomposed using a multi-scale wavelet, and the time domain amplitude change rate and energy distribution parameters of each frequency band are extracted to generate a multi-dimensional time-frequency domain feature vector;

[0009] The multidimensional time-frequency domain feature vector is calibrated by using an optical fiber synchronous transmission protocol, and a nonlinear manifold mapping is performed on the calibrated multidimensional time-frequency domain feature vector using a local linear embedding algorithm to generate a low-dimensional feature vector adapted to the dielectric spectrum characteristics;

[0010] The low-dimensional feature vector is input into a pre-built deep learning architecture, and an identification result including a classification probability distribution is output. The partial discharge type with the highest similarity to the identification result is matched from the historical discharge pattern database. The insulation degradation level is determined based on the final loss threshold interval in the dielectric spectrum characteristics, and a comprehensive diagnostic report including the partial discharge type and insulation degradation level is generated.

[0011] Optionally, the performing nonlinear manifold mapping on the calibrated multidimensional time-frequency domain feature vector based on a locally linear embedding algorithm to generate a low-dimensional feature vector adapted to the dielectric spectrum characteristics includes:

[0012] Determining a target local neighborhood radius corresponding to a multidimensional time-frequency domain feature vector based on a dielectric loss tangent value in the dielectric spectrum characteristics;

[0013] calculating, within the local neighborhood radius, a spatial distance weight between multidimensional time-frequency domain feature vectors, wherein the spatial distance weight is inversely proportional to a dispersion parameter in the dielectric spectrum characteristic;

[0014] Performing frequency band weighted correction on the spatial distance weight according to the frequency distribution in the dielectric spectrum characteristics to generate a corrected spatial distance weight;

[0015] Constructing a local linear reconstruction relationship between multidimensional time-frequency domain feature vectors using the modified spatial distance weights, and extracting basis vectors in the local linear reconstruction relationship;

[0016] Based on the absorption peak position in the dielectric spectrum characteristics, a low-dimensional feature vector corresponding to the absorption peak frequency band in the basis vector is retained.

[0017] Optionally, determining the target local neighborhood radius corresponding to the multidimensional time-frequency domain feature vector based on the dielectric loss tangent value in the dielectric spectrum characteristics includes:

[0018] Extracting dielectric loss tangent values ​​at different frequency points in the dielectric spectrum characteristics based on the spectrum characteristics of the partial discharge signal after interference is filtered out, so as to construct a dielectric loss tangent value sequence;

[0019] According to the dielectric loss tangent value sequence, an initial local neighborhood radius is obtained using a mapping table of the dielectric loss tangent value and the local neighborhood radius;

[0020] The initial local neighborhood radius is compensated and corrected according to the thickness of the high-voltage cable insulation layer to generate a target local neighborhood radius, which is used to limit the spatial neighborhood boundary of the multidimensional time-frequency domain feature vector.

[0021] Optionally, determining the insulation degradation level in combination with the final loss threshold interval in the dielectric spectrum characteristics includes:

[0022] Calculating the cumulative loss factor of the dielectric loss tangent value within the absorption peak frequency band;

[0023] Constructing an intermediate loss threshold interval based on the exponential decay relationship between the cumulative loss factor and the thickness of the high-voltage cable insulation layer, wherein the upper limit of the intermediate loss threshold interval is determined by the frequency slope of the absorption peak frequency band;

[0024] Performing frequency band sensitivity correction on the intermediate loss threshold interval based on the frequency dispersion parameter to generate a final loss threshold interval;

[0025] Mapping the energy distribution of the time-frequency domain feature vector corresponding to the partial discharge type to the final loss threshold interval to calculate the deviation of the energy of each frequency band from the corresponding final loss threshold interval;

[0026] The insulation degradation level is determined according to the cumulative proportion of the deviation within the absorption peak frequency band and the mapping relationship between the cumulative proportion and the insulation degradation level.

[0027] Optionally, constructing an intermediate loss threshold interval according to the exponential decay relationship between the cumulative loss factor and the thickness of the high-voltage cable insulation layer includes:

[0028] Extracting the maximum and minimum values ​​of the cumulative loss factor within the absorption peak frequency band, and calculating the logarithmic attenuation ratios of the maximum and minimum values ​​to the thickness of the high-voltage cable insulation layer;

[0029] Based on the logarithmic attenuation ratio and the frequency slope, constructing an exponential attenuation coefficient dynamic adjustment model;

[0030] Calculating an exponential attenuation coefficient corresponding to the thickness of the high-voltage cable insulation layer using the exponential attenuation coefficient dynamic adjustment model;

[0031] generating an initial loss threshold interval according to the exponential decay coefficient and the cumulative loss factor;

[0032] Based on the confidence interval of the thickness of the high-voltage cable insulation layer, error compensation is performed on the initial loss threshold interval to generate an intermediate loss threshold interval.

[0033] Optionally, the partial discharge signal in the high-voltage cable insulation layer is filtered out of power frequency and low-frequency interference through a Butterworth high-pass filter, the partial discharge signal after interference is filtered out is decomposed using a multi-scale wavelet, and the time domain amplitude change rate and energy distribution parameters of each frequency band are extracted to generate a multidimensional time-frequency domain feature vector, including:

[0034] Based on the lower limit of the sensor's operating frequency band and the frequency value corresponding to the minimum dielectric constant in the dielectric spectrum characteristics, the cutoff frequency threshold of the Butterworth high-pass filter is set to filter out power frequency and low-frequency interference;

[0035] Determining the number of levels of a multi-scale wavelet according to the number and intervals of absorption peak frequency bands in the dielectric spectrum characteristics, and using the multi-scale wavelet to perform hierarchical decomposition on the partial discharge signal after filtering out interference to obtain wavelet sub-band signals of each frequency band;

[0036] For the wavelet subband signals of each frequency band, extracting the time domain amplitude change rate of each frequency band to form an amplitude difference sequence;

[0037] Based on the attenuation gradient of the dielectric loss tangent value in the dielectric spectrum characteristics in the corresponding frequency band, performing frequency band weight correction on the time domain energy integral value of the wavelet subband signal of each frequency band to generate energy distribution parameters;

[0038] The amplitude difference sequence and the energy distribution parameters are processed in a manner corresponding to the scene to form a multi-dimensional time-frequency domain feature vector.

[0039] Optionally, the processing of the amplitude difference sequence and the energy distribution parameter in a manner corresponding to the scene to form a multidimensional time-frequency domain feature vector includes:

[0040] In an online monitoring scenario, the amplitude difference sequence and the energy distribution parameter are concatenated in frequency band order to form a multidimensional time-frequency domain feature vector;

[0041] In an offline monitoring scenario, the statistical variance and peak-to-peak ratio of the amplitude difference sequence are calculated, and the statistical variance, the peak-to-peak ratio and the energy distribution parameter are concatenated in frequency band order to form a multidimensional time-frequency domain feature vector.

[0042] In a second aspect, the present application provides a high-voltage cable multi-parameter intelligent diagnostic system, comprising:

[0043] An acquisition module is configured to deploy corresponding ultra-high frequency sensor arrays at the joints and terminals of the high-voltage cable to acquire partial discharge signals within the insulation layer of the high-voltage cable. The operating frequency band of the ultra-high frequency sensor array matches the dielectric spectrum characteristics of the high-voltage cable insulation material.

[0044] a decomposition module for filtering out power frequency and low-frequency interference from the partial discharge signal in the insulation layer of the high-voltage cable through a Butterworth high-pass filter, decomposing the partial discharge signal after interference removal using a multi-scale wavelet, and extracting the time-domain amplitude change rate and energy distribution parameters of each frequency band to generate a multi-dimensional time-frequency domain feature vector;

[0045] a calibration module for calibrating the multidimensional time-frequency domain feature vectors through an optical fiber synchronous transmission protocol, and performing nonlinear manifold mapping on the calibrated multidimensional time-frequency domain feature vectors using a locally linear embedding algorithm to generate low-dimensional feature vectors adapted to the dielectric spectrum characteristics;

[0046] A matching module is configured to input the low-dimensional feature vector into a pre-built deep learning architecture, output a recognition result including a classification probability distribution, match the partial discharge type with the highest similarity to the recognition result from a historical discharge pattern database, determine the insulation degradation level based on the final loss threshold interval in the dielectric spectrum characteristics, and generate a comprehensive diagnostic report including the partial discharge type and insulation degradation level.

[0047] In a third aspect, the present application provides a computing device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a high-voltage cable multi-parameter intelligent diagnosis method as described in any one of the first aspects.

[0048] In a fourth aspect, the present application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a high-voltage cable multi-parameter intelligent diagnosis method as described in any one of the first aspects.

[0049] In the present application, a multi-parameter intelligent diagnosis method for high-voltage cables is provided, which includes: deploying corresponding ultra-high frequency sensor arrays at the joints and terminal positions of the high-voltage cable respectively to obtain local discharge signals in the insulation layer of the high-voltage cable, wherein the operating frequency band of the ultra-high frequency sensor array matches the dielectric spectrum characteristics of the high-voltage cable insulation material; filtering the local discharge signals in the insulation layer of the high-voltage cable through a Butterworth high-pass filter to remove power frequency and low-frequency interference, using multi-scale wavelets to decompose the local discharge signals after filtering out interference, and extracting the time domain amplitude change rate and energy distribution parameters of each frequency band to generate a multi-dimensional time-frequency domain feature vector; The multidimensional time-frequency domain feature vector is calibrated through the optical fiber synchronous transmission protocol, and the calibrated multidimensional time-frequency domain feature vector is subjected to nonlinear manifold mapping based on a locally linear embedding algorithm to generate a low-dimensional feature vector adapted to the dielectric spectrum characteristics; the low-dimensional feature vector is input into a pre-built deep learning architecture, and an identification result including a classification probability distribution is output. The local discharge type with the highest similarity to the identification result is matched from a historical discharge pattern database, and the insulation degradation level is determined in combination with the final loss threshold interval in the dielectric spectrum characteristics, and a comprehensive diagnostic report including the local discharge type and the insulation degradation level is generated.

[0050] Beneficial effects of this application:

[0051] This application deploys an ultra-high frequency sensor array that matches the dielectric spectrum characteristics of the cable insulation material, combines a Butterworth high-pass filter to filter out power frequency interference and multi-scale wavelet decomposition, and constructs a multi-dimensional time-frequency feature vector; uses a fiber optic synchronization protocol to calibrate feature synchronization, and performs nonlinear manifold dimensionality reduction through a local linear embedding algorithm. Finally, based on deep learning classification and dynamic matching of dielectric loss thresholds, it achieves multi-dimensional and accurate diagnosis of the cable insulation status, thereby improving the accuracy and reliability of multi-parameter diagnosis of high-voltage cables.

[0052] Furthermore, the initial local neighborhood radius is determined by using the dielectric loss tangent value sequence and a preset mapping table, and the target neighborhood radius is obtained by combining the insulation layer thickness compensation correction. Within the target neighborhood, the spatial distance weight between feature vectors is calculated using the inverse relationship of the dielectric dispersion parameter. The weight is corrected by the frequency band distribution, and a local linear reconstruction relationship is constructed and basis vectors are extracted. Finally, based on the position of the dielectric absorption peak, low-dimensional feature vectors associated with the frequency band are selected to achieve deep coupling between the feature space and the dielectric properties. The neighborhood radius is dynamically calibrated by the dielectric loss tangent value to ensure that local structures strongly related to insulation degradation are retained during feature dimensionality reduction. The weight distribution is optimized based on the dispersion parameter and frequency band distribution to enhance the physical correlation between the feature space and the dielectric response characteristics. The absorption peak is combined with the basis vector to filter the redundant noise frequency band features, improve the accuracy of the low-dimensional features in representing the discharge mode and degradation degree, and provide high signal-to-noise ratio and strong interpretability input data for the subsequent classification model.

[0053] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0055] Figure 1 A flowchart of a multi-parameter intelligent diagnosis method for high-voltage cables provided in an embodiment of the present application;

[0056] Figure 2 A schematic diagram of the structure of a high-voltage cable multi-parameter intelligent diagnostic system provided in an embodiment of the present application;

[0057] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0058] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0059] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 11, 12, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0060] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0061] In order to solve the problems in the prior art of signal distortion caused by mismatch between fixed-frequency sensors and the dielectric spectrum of insulating materials, insufficient robustness of feature extraction caused by modal aliasing in empirical mode decomposition, and lack of ability of principal component analysis dimensionality reduction to characterize the nonlinear manifold structure of discharge signals, which ultimately lead to low accuracy and poor reliability of multi-parameter diagnosis of high-voltage cables, the embodiment of the present application provides a multi-parameter intelligent diagnosis method and system for high-voltage cables. The method adopts the following concept: first, the frequency band of the ultra-high frequency sensor array is customized according to the dielectric spectrum of the insulating material, the power frequency interference is suppressed by a Butterworth high-pass filter and combined with multi-scale wavelet decomposition, and multi-dimensional time-frequency features are constructed to characterize the differences in discharge modes; then, the fiber optic synchronization protocol is used to eliminate the time delay of multi-sensor data, and a local linear embedding algorithm driven by dielectric properties is adopted to form low-dimensional and high-discriminative features; finally, a deep learning model is used to fuse the probability classification of discharge modes with dynamic matching of dielectric loss thresholds to achieve multi-dimensional interpretable diagnosis of cable insulation degradation, forming a closed-loop optimization link from signal perception to state assessment.

[0062] Figure 1 A flowchart of a multi-parameter intelligent diagnosis method for high-voltage cables provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes:

[0063] S11. Deploy corresponding ultra-high frequency sensor arrays at the joints and terminals of the high-voltage cable to obtain partial discharge signals in the insulation layer of the high-voltage cable. The operating frequency band of the ultra-high frequency sensor array matches the dielectric spectrum characteristics of the high-voltage cable insulation material.

[0064] Among them, the ultra-high frequency sensor array refers to an array composed of multiple ultra-high frequency electromagnetic wave sensors, which is used to detect cable local discharge signals. The high-voltage cable insulation layer refers to the insulating material layer used to isolate the conductor in the cable, such as cross-linked polyethylene. The local discharge signal refers to the transient electromagnetic pulse signal generated by the local breakdown caused by the concentration of the electric field inside the insulating material. The working frequency band refers to the effective signal receiving frequency range designed by the sensor. Commonly used insulating materials for high-voltage cables mainly include cross-linked polyethylene, ethylene propylene rubber, silicone rubber, polyvinyl chloride, etc. The dielectric spectrum characteristics include dielectric constant, dielectric loss tangent value, and absorption peak frequency band. Different node spectrum characteristics can correspond to different working frequency bands, and the setting of the corresponding relationship between the two can be set according to expert experience or adaptive setting method. The embodiments of this application do not make specific limitations on this.

[0065] In this embodiment, an array of ultra-high frequency (UHF) sensors is deployed at the connectors and terminals of a high-voltage cable. The array's operating frequency band is designed to match the dielectric spectrum characteristics of the cable's insulation material, ensuring the sensors can efficiently capture the UHF electromagnetic wave signals generated by partial discharge within the cable's insulation. This frequency band matching ensures that the sensor array only receives partial discharge signals that are compatible with the insulation material's dielectric response characteristics, avoiding signal attenuation or noise interference caused by frequency band mismatch.

[0066] S12. The local discharge signal in the insulation layer of the high-voltage cable is filtered out of power frequency and low-frequency interference through a Butterworth high-pass filter. The local discharge signal after interference filtering is decomposed using a multi-scale wavelet, and the time domain amplitude change rate and energy distribution parameters of each frequency band are extracted to generate a multi-dimensional time-frequency domain feature vector.

[0067] The Butterworth high-pass filter is a ripple-free high-pass filter used to filter out low-frequency signals. Power frequency refers to the operating frequency of the power system. Low-frequency interference refers to noise signals with frequencies lower than those of partial discharge signals. Multiscale wavelets decompose signals using wavelet basis functions of different scales to extract multi-resolution features. The time-domain amplitude change rate refers to the rate at which the signal amplitude changes over time. The energy distribution parameter refers to the proportion of signal energy in each frequency band to the total energy. A multidimensional time-frequency domain feature vector is a multidimensional data vector that integrates time-domain and frequency-domain features.

[0068] In an embodiment of the present application, the acquired partial discharge signal is input into a Butterworth high-pass filter to filter out the power frequency and low-frequency interference signals and retain the high-frequency partial discharge components; then the filtered signal is subjected to multi-scale wavelet decomposition, the frequency bands are divided according to a preset scale, and the time domain amplitude change rate and energy distribution parameters of the signal in each frequency band are extracted respectively, and a multidimensional time-frequency domain feature vector containing time-frequency domain features is generated by fusion to characterize the pattern differences of the discharge signal.

[0069] S13. The multidimensional time-frequency domain feature vector is calibrated through the optical fiber synchronous transmission protocol, and the calibrated multidimensional time-frequency domain feature vector is subjected to nonlinear manifold mapping based on a local linear embedding algorithm to generate a low-dimensional feature vector adapted to the dielectric spectrum characteristics.

[0070] Among them, the Fiber Synchronous Transmission Protocol is a communication protocol that transmits and synchronizes multiple signals through optical fibers. The local linear embedding algorithm is a nonlinear dimensionality reduction algorithm used to maintain local neighborhood structure. Nonlinear manifold mapping is a method of mapping high-dimensional data to a low-dimensional manifold space. The dimension of the low-dimensional feature vector is usually determined by the physical correlation of the dielectric spectrum characteristics and the requirements of the classification model. In practical applications, it is generally 3-8 dimensions. The low-dimensional feature vector is not a simple compressed random parameter, but a potential physical feature deeply associated with the dielectric spectrum characteristics extracted by the local linear embedding algorithm.

[0071] In an embodiment of the present application, the generated multi-dimensional time-frequency domain feature vectors are time-synchronized and calibrated through the optical fiber synchronous transmission protocol to eliminate the transmission delay error between the multi-sensor arrays; then a local linear embedding algorithm is used to construct a nonlinear manifold mapping model based on the local neighborhood relationship between the feature vectors, and the high-dimensional multi-dimensional time-frequency domain feature vectors are mapped to a low-dimensional space adapted to the dielectric spectrum characteristics, retaining the potential structural characteristics associated with the dielectric characteristics, and generating a low-dimensional feature vector.

[0072] S14. Input the low-dimensional feature vector into a pre-built deep learning architecture, output the recognition result including the classification probability distribution, match the partial discharge type with the highest similarity to the recognition result from the historical discharge pattern database, and determine the insulation degradation level in combination with the final loss threshold interval in the dielectric spectrum characteristics, and generate a comprehensive diagnostic report including the partial discharge type and insulation degradation level.

[0073] Among them, the deep learning architecture is a machine learning model based on a neural network. The classification probability distribution refers to the set of probability values ​​for each category output by the model. The recognition result is the partial discharge type determined by the model. The historical discharge pattern database refers to a database that stores characteristic data of known discharge types. The partial discharge type refers to the discharge pattern classification result, including air gap discharge, surface discharge, corona discharge, surface discharge, and tree discharge. The final loss threshold interval is the insulation degradation level threshold range defined based on dielectric loss. The insulation degradation level refers to the classification of the degree of aging of the insulation material, such as mild, moderate, and severe. The comprehensive diagnostic report is an assessment result document that includes the discharge type and degradation level.

[0074] In an embodiment of the present application, the obtained low-dimensional feature vector is input into a pre-trained deep learning architecture, and the classification probability distribution of different partial discharge types is output through forward propagation of the model. Based on the probability distribution, the partial discharge type with the highest similarity is matched from the historical discharge pattern database, and combined with the final loss threshold interval preset in the dielectric spectrum characteristics, the insulation degradation level is dynamically determined, and finally a comprehensive diagnostic report including the discharge type and degradation level is generated.

[0075] The following is a specific example: Taking the monitoring of a certain high-voltage cable line as an example, first, an ultra-high frequency sensor array with an operating frequency band of 1.2-1.8 GHz is deployed at the cable joint to collect local discharge pulse signals in real time; the original signal is filtered out of power frequency interference using a Butterworth high-pass filter, and a five-layer decomposition is performed using multi-scale wavelets to extract the time domain amplitude change rate and energy distribution parameters of each frequency band to generate a 16-dimensional time-frequency feature vector; after aligning the timestamps of the multi-sensor feature vectors using the fiber synchronization protocol, the feature dimension is reduced to 3 dimensions using a local linear embedding algorithm, retaining the characteristic components associated with the dielectric absorption peak; finally, the low-dimensional features are input into a pre-trained deep learning architecture, which outputs an air gap discharge probability of 85% and a surface discharge probability of 12%. Combined with the dielectric loss threshold range, a comprehensive diagnostic report of "air gap discharge, severe insulation degradation level" is generated to guide operation and maintenance personnel to urgently replace the cable joints, realizing a full-link closed-loop application from signal acquisition to fault decision-making.

[0076] By executing S11 to S14, the embodiment of the present application accurately captures local discharge signals through a frequency-band-matched ultra-high frequency sensor array, combines Butterworth filtering with multi-scale wavelet decomposition to suppress interference and extract high-resolution time-frequency features; utilizes the optical fiber synchronization protocol and the local linear embedding algorithm to achieve feature calibration and non-linear physical correlation dimensionality reduction, and finally dynamically evaluates the insulation status through the fusion of the dielectric loss threshold through the deep learning model, thereby improving the accuracy of local discharge classification and the reliability of degradation level determination, and providing a highly reliable diagnostic basis for cable operation and maintenance.

[0077] In a possible embodiment, S13, performing nonlinear manifold mapping on the calibrated multidimensional time-frequency domain feature vector using a locally linear embedding algorithm to generate a low-dimensional feature vector adapted to the dielectric spectrum characteristics, includes:

[0078] Step 131: Determine a target local neighborhood radius corresponding to a multi-dimensional time-frequency domain feature vector based on a dielectric loss tangent value in the dielectric spectrum characteristics.

[0079] The dielectric loss tangent is a parameter that characterizes the energy loss of insulating materials and is defined as the ratio of the dielectric loss factor to the imaginary part of the dielectric constant. The target local neighborhood radius is the neighborhood range corrected for dielectric properties and material thickness and is used to define the local spatial boundaries of the eigenvector. The multidimensional time-frequency domain eigenvector is a joint time-frequency feature extracted from the partial discharge signal, reflecting the correlation between the pattern differences of the discharge signal and the dielectric properties of the insulating material.

[0080] In an embodiment of the present application, first, based on the dielectric spectrum characteristics of the high-voltage cable insulation material, the values ​​of its dielectric loss tangent at different frequency points are extracted to form a dielectric loss tangent value sequence; according to a preset mapping table of dielectric loss tangent and local neighborhood radius, the initial local neighborhood radius is obtained; the initial neighborhood radius is further compensated and corrected in combination with the thickness of the high-voltage cable insulation layer to generate a target local neighborhood radius, which is used to limit the spatial neighborhood boundary of the multi-dimensional time-frequency domain feature vector to ensure that the local structural features that are strongly correlated with insulation degradation are retained during dimensionality reduction.

[0081] Step 132: Calculate the spatial distance weight between the multi-dimensional time-frequency domain feature vectors within the local neighborhood radius. The spatial distance weight is inversely proportional to the dispersion parameter in the dielectric spectrum characteristics.

[0082] The spatial distance weight is the weighted coefficient of the distance between eigenvectors, reflecting the influence of dispersion on feature similarity. The dispersion parameter is the rate at which the dielectric constant changes with frequency and is used to quantify the dispersion characteristics of a material.

[0083] In an embodiment of the present application, within a determined target local neighborhood radius, the Euclidean space distance between multidimensional time-frequency domain feature vectors is calculated, and the dispersion parameter in the dielectric spectrum characteristics is used as a weight adjustment factor. The spatial distance weight is defined as the inverse of the dispersion parameter, thereby suppressing the characteristic distortion of the high-frequency band signal due to the dispersion effect and enhancing the contribution of the low-frequency band features.

[0084] Step 133: Perform frequency band weighting correction on the spatial distance weight according to the frequency distribution in the dielectric spectrum characteristics to generate a corrected spatial distance weight.

[0085] Among them, the corrected spatial distance weight refers to the weight adjusted in combination with the frequency distribution, which strengthens the importance of high-energy frequency band characteristics.

[0086] In an embodiment of the present application, the spatial distance weight of each frequency band is multiplied by the frequency distribution coefficient of the frequency band to generate a corrected spatial distance weight, so that the features of the high-energy frequency band obtain higher weights in the reconstruction relationship, thereby strengthening the physical correlation of the feature space.

[0087] Step 134: Use the modified spatial distance weights to construct a local linear reconstruction relationship between the multi-dimensional time-frequency domain feature vectors, and extract basis vectors in the local linear reconstruction relationship.

[0088] Among them, the local linear reconstruction relationship refers to the linear combination relationship based on the eigenvectors in the neighborhood, which is used to describe the local manifold structure. The basis vector refers to the principal component vector solved by linear reconstruction, which can characterize the key mode of the feature space. The embodiment of the present application is based on the modified spatial distance weight, and the neighborhood range of the multi-dimensional time-frequency domain eigenvector is delineated with the target local neighborhood radius; the local linear relationship is constructed by minimizing the weighted reconstruction error, and the closed-form solution that satisfies the weight sum of 1 is solved, for example, the weights of the three points in the neighborhood are assigned to [0.6, 0.3, 0.1]; then the global weight matrix is ​​spliced ​​and singular value decomposition is performed, and the first three principal components are extracted as basis vectors, for example, basis vector 1 represents the energy correlation characteristics of the 1.5GHz frequency band, and basis vector 2 reflects the time domain mutation and dispersion coupling characteristics; finally, the dielectric absorption peak position is combined to screen the basis vectors that are strongly correlated with the absorption peak frequency band to provide data support for the generation of low-dimensional feature vectors, which can achieve physically driven high-dimensional feature compression and focusing of insulation degradation sensitive information.

[0089] In an embodiment of the present application, the generated corrected spatial distance weight is used to construct a local linear reconstruction relationship between multidimensional time-frequency domain eigenvectors, that is, by minimizing the linear combination error of the eigenvectors in the neighborhood, the basis vectors representing the local manifold structure are solved, and the basis vectors reflect the key patterns associated with the dielectric properties in the feature space.

[0090] Step 135 : Based on the absorption peak position in the dielectric spectrum characteristics, retain the low-dimensional feature vector corresponding to the absorption peak frequency band in the basis vector.

[0091] The absorption peak position refers to the frequency point where energy is absorbed in the dielectric spectrum, and the absorption peak frequency band refers to the specific frequency range surrounding the absorption peak position.

[0092] In an embodiment of the present application, based on the absorption peak position in the dielectric spectrum characteristics, the components corresponding to the absorption peak frequency band in the extracted basis vector are screened, irrelevant frequency band features are eliminated, and the components in the low-dimensional feature vector that are strongly correlated with the dielectric response of the insulating material are retained to avoid noise interference and improve feature discriminability.

[0093] The following is a specific example: Taking a high-voltage cable monitoring scenario as an example, the dielectric loss tangent value sequence of its dielectric spectrum in the 1.2-1.8 GHz frequency band is first extracted, and the initial local neighborhood radius of 0.6 is obtained according to the preset mapping table. The compensation correction is based on the insulation layer thickness of 5 mm to obtain the target neighborhood radius of 0.5; the Euclidean distance between the multidimensional time-frequency domain feature vectors is calculated in the neighborhood of radius 0.5, and the spatial distance weight of 1.25 is generated by combining the dispersion parameter of 0.8 at 1.5 GHz; the weight is frequency-band weighted correction is performed based on the 40% energy share of the 1.5 GHz frequency band, and the corrected weight is 0.5; the corrected weight is used to construct a local linear reconstruction relationship, solve the basis vector, and extract the first and third principal components; finally, the components in the basis vector corresponding to the 1.4-1.6 GHz absorption peak frequency band are filtered to generate a three-dimensional low-dimensional feature vector, which is input into the classification model to achieve accurate assessment of insulation degradation.

[0094] By executing steps 131 to 135, the embodiment of the present application dynamically calibrates the neighborhood radius through the dielectric loss tangent value, combines the dispersion parameter and the frequency distribution to optimize the spatial weight distribution, constructs a physically associated local linear reconstruction relationship, and screens the basis vectors based on the absorption peak to achieve deep coupling of feature dimensionality reduction and dielectric properties, effectively remove redundant noise, enhance feature discriminability, and provide low-dimensional feature input with high signal-to-noise ratio and strong interpretability for insulation degradation assessment.

[0095] In a possible embodiment, step 131, determining a target local neighborhood radius corresponding to a multi-dimensional time-frequency domain feature vector based on a dielectric loss tangent value in a dielectric spectrum characteristic, includes:

[0096] Step a1: based on the spectrum characteristics of the partial discharge signal after interference is filtered out, the dielectric loss tangent values ​​at different frequency points in the dielectric spectrum characteristics are extracted to construct a dielectric loss tangent value sequence.

[0097] Spectral characteristics refer to the amplitude, phase, or energy distribution characteristics of a signal in the frequency domain, reflecting the contributions of different frequency components. A dielectric loss tangent sequence is a collection of dielectric loss tangent values ​​arranged in frequency order, characterizing the energy loss of a material at different frequencies.

[0098] In an embodiment of the present application, based on the spectral characteristics of the local discharge signal after filtering out interference, the dielectric loss tangent values ​​at different frequency points in the dielectric spectrum characteristics of the high-voltage cable insulation material are extracted through Fourier transform or power spectral density analysis, and a dielectric loss tangent value sequence is constructed by arranging the frequency in ascending order with the frequency as the horizontal axis and the dielectric loss tangent value as the vertical axis, which is used to quantify the energy loss characteristics of the insulation material in different frequency bands.

[0099] Step a2: According to the dielectric loss tangent value sequence, an initial local neighborhood radius is obtained using a mapping table of dielectric loss tangent values ​​and local neighborhood radius.

[0100] The initial local neighborhood radius refers to the initial value of the neighborhood range obtained from the mapping table based on the dielectric loss characteristics, and is used to define the local spatial range of the feature vector.

[0101] In an embodiment of the present application, based on the generated sequence of dielectric loss tangent values, a preset mapping table of dielectric loss tangent values ​​and local neighborhood radius is queried. The mapping table can be obtained based on experimental calibration, and the initial local neighborhood radius value corresponding to each frequency point is determined through linear interpolation or nearest neighbor matching algorithm to form an initial neighborhood radius set dynamically associated with the dielectric loss characteristics, providing benchmark parameters for subsequent spatial neighborhood boundary delineation.

[0102] Step a3: Compensate and correct the initial local neighborhood radius according to the thickness of the high-voltage cable insulation layer to generate a target local neighborhood radius. The target local neighborhood radius is used to limit the spatial neighborhood boundary of the multidimensional time-frequency domain feature vector. The unit can be millimeters or other units. This application does not make specific restrictions on this.

[0103] The thickness of the high-voltage cable insulation layer refers to the physical thickness of the cable insulation material layer, which affects signal attenuation and electric field distribution. The spatial neighborhood boundary refers to the geometric boundary that defines the local neighborhood range of the feature vector and is used to constrain the calculation area of ​​the dimensionality reduction algorithm.

[0104] In an embodiment of the present application, the initial local neighborhood radius obtained is compensated and corrected according to the thickness of the high-voltage cable insulation layer: the initial radius is multiplied by the thickness compensation coefficient to generate a target local neighborhood radius, which is used to limit the spatial neighborhood boundary of the multi-dimensional time-frequency domain feature vector to ensure that the local structural features that are strongly correlated with the insulation thickness and dielectric loss are retained during feature dimensionality reduction.

[0105] The following is a specific example: Taking the monitoring of a cable joint as an example, the spectrum analysis of the local discharge signal after interference filtering is first performed, and the dielectric loss tangent value at every 100 MHz interval in the 1.0-2.0 GHz frequency band is extracted to construct a dielectric loss tangent value sequence of [0.012, 0.015, 0.018]. Then, according to the preset mapping table, the initial local neighborhood radius of 0.55 mm corresponding to the dielectric loss tangent value of 0.018 at 1.5 GHz is calculated by linear interpolation. Finally, based on the cable insulation layer thickness of 8 mm, the compensation coefficient formula is used to calculate the correction factor of 6.6, and the initial radius of 0.55 mm is corrected to the target local neighborhood radius of 3.63 mm. The spatial neighborhood boundary of the multidimensional time-frequency domain feature vector is delineated as a radius of 3.63 mm for subsequent local manifold structure analysis and feature dimensionality reduction.

[0106] By executing steps a1 to a3, the embodiment of the present application dynamically maps the initial neighborhood radius through a sequence of dielectric loss tangent values, and combines it with insulation layer thickness compensation correction to achieve precise adaptation of the feature space neighborhood boundary and the physical properties of the material, avoiding feature distortion or information loss caused by a fixed neighborhood radius, and providing local structural constraints strongly associated with the insulation state for subsequent nonlinear dimensionality reduction.

[0107] In a possible embodiment, S14, determining the insulation degradation level in combination with the final loss threshold interval in the dielectric spectrum characteristics, includes:

[0108] Step 141: Calculate the cumulative loss factor of the dielectric loss tangent within the absorption peak frequency band.

[0109] Among them, the cumulative loss factor refers to the cumulative sum of the dielectric loss tangent values ​​​​of all frequency points in the absorption peak frequency band, which represents the overall energy loss in this frequency band.

[0110] In an embodiment of the present application, the dielectric loss tangent values ​​at all frequency points within the absorption peak frequency band are first extracted, and the cumulative loss factor is generated by accumulating the dielectric loss tangent values ​​at each frequency point within the frequency band to quantify the overall energy loss intensity of the absorption peak frequency band.

[0111] Step 142: construct an intermediate loss threshold interval based on the exponential decay relationship between the cumulative loss factor and the thickness of the high-voltage cable insulation layer. The upper limit of the intermediate loss threshold interval is determined by the frequency slope of the absorption peak frequency band.

[0112] Among them, the exponential decay relationship refers to the mathematical model between the thickness of the insulation layer and the loss threshold, and the formula is The intermediate loss threshold interval is the threshold range initially determined by exponential decay and frequency slope, used for energy deviation calculation. The frequency slope is the derivative of the dielectric loss tangent as it changes with frequency, reflecting the sensitivity of the loss to frequency.

[0113] In an embodiment of the present application, the initial range of the intermediate loss threshold interval is calculated based on the exponential attenuation relationship between the cumulative loss factor and the thickness of the high-voltage cable insulation layer; the upper limit value is further adjusted by the frequency slope of the absorption peak frequency band, and the calculation formula for the upper limit of the intermediate loss threshold interval is: upper limit = initial upper limit × frequency slope.

[0114] Step 143: Perform frequency band sensitivity correction on the intermediate loss threshold interval based on the dispersion parameter to generate a final loss threshold interval.

[0115] Frequency band sensitivity correction refers to adjusting the threshold range based on the dispersion parameter to increase the judgment weight of highly sensitive frequency bands. The final loss threshold range refers to the corrected dynamic threshold range, which is used to quantify the degree of energy distribution anomaly.

[0116] In an embodiment of the present application, the upper and lower limits of the intermediate threshold interval are multiplied by the normalized value of the dispersion parameter to generate a final loss threshold interval to reflect the difference in sensitivity of different frequency bands to insulation degradation.

[0117] Step 144 : Map the energy distribution of the time-frequency domain feature vector corresponding to the partial discharge type to the final loss threshold interval to calculate the deviation of the energy of each frequency band from the corresponding final loss threshold interval.

[0118] The energy distribution of the time-frequency domain eigenvector refers to the energy proportion of each frequency band after multi-scale wavelet decomposition. Band energy refers to the ratio of the signal energy in a specific frequency band to the total energy. The deviation from the final loss threshold interval refers to the relative deviation between the frequency band energy value and the threshold interval, which is used to quantify the degree of anomaly.

[0119] In the embodiment of the present application, if the frequency band energy is higher than the upper threshold, the deviation amount = energy value / upper threshold value-1; if it is lower than the lower limit, the deviation amount = 1-energy value / lower threshold value. In other cases, the deviation amount is 0.

[0120] Step 145 : Determine the insulation degradation level based on the cumulative proportion of the deviation within the absorption peak frequency band and the mapping relationship between the cumulative proportion and the insulation degradation level.

[0121] In this embodiment of the present application, the cumulative percentage of all deviations within the absorption peak frequency band is calculated, and the insulation degradation level is determined based on a preset mapping relationship between the cumulative percentage and the insulation degradation level. The "cumulative percentage" refers to the percentage of the sum of all deviations within the absorption peak frequency band to the total deviations across the entire frequency band.

[0122] The following is a specific example: Taking the monitoring of a high-voltage cable joint as an example, the dielectric loss tangent value in the 1.4-1.6GHz absorption peak frequency band is first extracted, and the cumulative loss factor is calculated to be 0.063; then, based on the cable insulation layer thickness of 8mm and the exponential decay formula, the upper limit of the intermediate threshold is calculated to be ≈0.028. Combined with the frequency slope of the absorption peak frequency band of 0.005 GHz, the upper limit can be corrected to 0.00014, generating the intermediate loss threshold interval [0.00014, 0.015]. Then, based on the dispersion parameter 0.8, the interval is sensitivity-corrected to obtain the final threshold interval [0.000112, 0.012]. The energy proportions of the 1.5 GHz and 1.4 GHz frequency bands are mapped to this interval, and the calculated deviations are 32.33 and 24, respectively. Finally, the cumulative deviation of the absorption peak frequency band is statistically calculated to account for 85%. According to the preset mapping relationship, the insulation degradation level is determined to be "severely degraded", realizing accurate status assessment driven by dynamic thresholds.

[0123] By executing steps 141 to 145, the embodiment of the present application dynamically constructs a threshold range through the cumulative loss factor, combines the dispersion parameter correction and energy deviation calculation, and realizes a deep correlation between insulation degradation assessment and dielectric properties, breaking through the limitations of traditional fixed thresholds, and improving the sensitivity and reliability of degradation level determination, which is particularly suitable for cable status diagnosis under complex frequency-varying characteristics.

[0124] In a possible embodiment, step 142, constructing an intermediate loss threshold interval based on an exponential decay relationship between a cumulative loss factor and a thickness of a high-voltage cable insulation layer, includes:

[0125] Step b1: extract the maximum and minimum values ​​of the cumulative loss factor within the absorption peak frequency band, and calculate the logarithmic attenuation ratios of the maximum and minimum values ​​to the thickness of the high-voltage cable insulation layer.

[0126] The maximum and minimum values ​​within the absorption peak frequency band refer to the highest and lowest values ​​of the dielectric loss tangent within the specified frequency band. The logarithmic attenuation ratio is the ratio of the loss factor to the natural logarithm of the thickness, reflecting the impact of thickness on loss attenuation.

[0127] In an embodiment of the present application, the maximum and minimum values ​​are first extracted from the cumulative loss factor sequence of the absorption peak frequency band, and the logarithmic attenuation ratios thereof to the thickness of the high-voltage cable insulation layer are calculated respectively: the logarithmic attenuation ratio corresponding to the maximum value is the maximum value divided by the natural logarithm of the thickness, and the logarithmic attenuation ratio corresponding to the minimum value is the minimum value divided by the natural logarithm of the thickness, which is used to quantify the attenuation rate of the loss factor as the thickness changes.

[0128] Step b2: Based on the logarithmic attenuation ratio and the frequency slope, a dynamic adjustment model of the exponential attenuation coefficient is constructed.

[0129] The exponential attenuation coefficient dynamic adjustment model refers to a mathematical model that integrates the frequency slope and the logarithmic attenuation ratio, and is used to dynamically correct the attenuation coefficient.

[0130] In an embodiment of the present application, based on the obtained logarithmic attenuation ratio and the frequency slope of the absorption peak frequency band, a dynamic adjustment model of the exponential attenuation coefficient is constructed: the frequency slope is added to the logarithmic attenuation ratio, and the dynamic adjustment coefficient is generated through normalization processing.

[0131] Step b3: Calculate the exponential attenuation coefficient corresponding to the thickness of the high-voltage cable insulation layer using the exponential attenuation coefficient dynamic adjustment model.

[0132] Among them, the exponential decay coefficient is a parameter that characterizes the exponential decay rate of the loss threshold with thickness and is calculated by the model.

[0133] In an embodiment of the present application, the generated exponential attenuation coefficient dynamic adjustment model is used in combination with the thickness of the high-voltage cable insulation layer to calculate the actual exponential attenuation coefficient: the thickness is multiplied by the adjustment coefficient, and then mapped to the final attenuation coefficient through an exponential function to achieve a nonlinear correlation between thickness and loss threshold.

[0134] Step b4: Generate an initial loss threshold interval based on the exponential decay coefficient and the cumulative loss factor.

[0135] The initial loss threshold range is a threshold range preliminarily determined based on the attenuation coefficient and cumulative loss factor. The product of the exponential attenuation coefficient and the cumulative loss factor determines the lower limit of the range, while the upper limit is expanded by introducing a frequency slope adjustment factor to form the range.

[0136] In the embodiment of the present application, the lower limit of the initial loss threshold interval = cumulative loss factor × attenuation coefficient, and the upper limit of the initial loss threshold interval = lower limit × frequency slope adjustment factor.

[0137] Step b5: Based on the confidence interval of the thickness of the high-voltage cable insulation layer, perform error compensation on the initial loss threshold interval to generate an intermediate loss threshold interval.

[0138] The confidence interval is the statistical error range of the thickness measurement and is used to quantify uncertainty. Error compensation is to expand the threshold range based on the confidence interval to offset the impact of measurement error.

[0139] In the embodiment of the present application, the larger the average confidence value corresponding to the confidence interval, the smaller the error compensation amount, and the influence of thickness measurement uncertainty can be eliminated based on step b5.

[0140] The following is a specific example: Taking the monitoring of a cable joint as an example, the maximum value of the dielectric loss tangent within the 1.4-1.6 GHz absorption peak frequency band is first extracted, which is 0.025, and the minimum value of 0.015. The logarithmic attenuation ratios of these values ​​to the thickness are calculated to be 0.012 and 0.0072, respectively. Then, combined with the frequency slope of the absorption peak frequency band of 0.005 GHz, a dynamic adjustment model for the exponential attenuation coefficient is constructed, and the dynamic adjustment coefficient is calculated to be 0.85. This coefficient is used to calculate the exponential attenuation coefficient exp(-(8×0.85) / 10)=0.503, and combined with the cumulative loss factor of 0.063 to generate the initial loss threshold interval [0.032, 0.048]. Finally, a 10% error compensation is performed based on the thickness confidence interval [7.5 mm, 8.5 mm], and the initial interval is expanded to the intermediate loss threshold interval [0.035, 0.053] to eliminate the influence of thickness measurement uncertainty on threshold determination and improve the robustness of degradation assessment.

[0141] By executing steps b1 to b5, the embodiment of the present application dynamically adjusts the exponential attenuation coefficient and the error compensation mechanism, integrates the insulation layer thickness, frequency slope, and measurement error into the threshold interval generation process, breaks through the limitations of the traditional static threshold, improves the adaptability of the loss threshold to changes in material properties and operating conditions, and enhances the robustness and reliability of degradation assessment.

[0142] In one possible embodiment, S12, the local discharge signal in the high-voltage cable insulation layer is filtered out of power frequency and low-frequency interference through a Butterworth high-pass filter, the local discharge signal after the interference is filtered out is decomposed using a multi-scale wavelet, and the time domain amplitude change rate and energy distribution parameters of each frequency band are extracted to generate a multi-dimensional time-frequency domain feature vector, including:

[0143] Step 121 : Based on the lower limit of the sensor operating frequency band and the frequency value corresponding to the minimum dielectric constant in the dielectric spectrum characteristics, set the cutoff frequency threshold of the Butterworth high-pass filter to filter out power frequency and low-frequency interference.

[0144] The lower limit of the sensor's operating frequency band refers to the lowest frequency at which the sensor can effectively receive signals. This is determined by the dielectric constant and thickness of the material. The minimum dielectric constant refers to the minimum relative dielectric constant of the insulating material within a specific frequency band and affects the cutoff frequency for electromagnetic wave propagation. The cutoff frequency threshold is the lowest effective frequency for electromagnetic waves to propagate through the insulating layer.

[0145] Step 122: Determine the number of levels of the multi-scale wavelet according to the number and intervals of absorption peak frequency bands in the dielectric spectrum characteristics, and use the multi-scale wavelet to perform hierarchical decomposition on the partial discharge signal after filtering out interference to obtain wavelet sub-band signals of each frequency band.

[0146] The number of absorption peak frequency bands refers to the number of frequency bands with energy absorption peaks in the dielectric spectrum. The frequency band spacing refers to the width of the frequency bands used for signal decomposition. Hierarchical decomposition involves decomposing the signal into subband signals layer by layer, from high to low frequency, using a multi-scale wavelet transform. The wavelet subband signal for each frequency band is the sequence of detail coefficients corresponding to each frequency band after wavelet decomposition.

[0147] Step 123 : For the wavelet sub-band signals of each frequency band, extract the time domain amplitude change rate of each frequency band to form an amplitude difference sequence.

[0148] The time-domain amplitude change rate refers to the rate at which the signal amplitude changes over time, calculated by the difference between adjacent points. The amplitude difference sequence refers to the sequence of amplitude differences between adjacent sampling points of the signal, reflecting the mutation characteristics.

[0149] Step 124 : Based on the attenuation gradient of the dielectric loss tangent value in the dielectric spectrum characteristics in the corresponding frequency band, perform frequency band weight correction on the time domain energy integral value of the wavelet subband signal in each frequency band to generate energy distribution parameters.

[0150] The attenuation gradient is the ratio of the mean amplitude difference to the subband energy, representing the strength of signal mutations. The time-domain energy integral is the sum of the squared signal amplitudes in the time domain, quantifying the energy of the frequency band. The energy distribution parameter is the ratio of the energy integral of each frequency band to the total energy, characterizing the energy distribution.

[0151] Step 125: Process the amplitude difference sequence and the energy distribution parameters in a manner corresponding to the scene to form a multi-dimensional time-frequency domain feature vector.

[0152] The following is a specific example: Taking a cable monitoring as an example, the cutoff frequency threshold is calculated as 3×10 based on the minimum dielectric constant and thickness of the insulation material. 8 Divide by 2×3.14 multiply by 2.3 multiply by 0.005, the result is about 1.2GHz, and set the Butterworth high-pass filter cutoff frequency to 1.2GHz to filter out power frequency and low-frequency interference; then detect the three absorption peak frequency bands of 1.2GHz, 1.5GHz, and 1.8GHz, perform 5-layer multi-scale wavelet decomposition at 200MHz intervals, generate the first wavelet subband signal covering 1.6-3.2GHz, the second wavelet subband signal covering 0.8-1.6GHz and other frequency bands; calculate the time domain amplitude difference sequence for the D2 subband signal, and extract the mean The original energy integral value 12.3V² is weighted 5% corrected based on the attenuation gradient of 0.005 / GHz of the dielectric loss tangent value in the 1.5GHz frequency band to obtain a corrected energy of 12.9V², and its proportion to the total energy of 86.5V² in the entire frequency band is calculated to be 14.9%; finally, the mean, variance and energy distribution parameters of the second wavelet subband signal are sequentially spliced ​​with other frequency band features to form a 16-dimensional time-frequency domain feature vector, which is input into the convolutional neural network model to realize partial discharge mode classification and insulation degradation assessment.

[0153] By executing steps 121 to 125, the embodiment of the present application dynamically constrains the sensor operating frequency band through a cutoff frequency threshold, combines multi-scale wavelet hierarchical decomposition and attenuation gradient analysis, and realizes refined time-frequency feature analysis of partial discharge signals, effectively distinguishes energy distribution and mutation patterns in different frequency bands, and provides a high-resolution physical correlation feature set for insulation degradation assessment.

[0154] In a possible embodiment, step 125 processes the amplitude difference sequence and the energy distribution parameter in a manner corresponding to the scene to form a multi-dimensional time-frequency domain feature vector, including:

[0155] Step c1: In an online monitoring scenario, the amplitude difference sequence and the energy distribution parameters are concatenated in frequency band order to form a multi-dimensional time-frequency domain feature vector.

[0156] Among them, the online monitoring scenario refers to a mode of collecting and processing data in real time, which requires low latency and high efficiency.

[0157] Step c2: In the offline monitoring scenario, the statistical variance and peak-to-peak ratio of the amplitude difference sequence are calculated, and the statistical variance, peak-to-peak ratio and energy distribution parameters are concatenated in the order of frequency bands to form a multidimensional time-frequency domain feature vector.

[0158] The offline monitoring scenario batch processing mode for historical data allows for complex calculations and high-precision analysis. Statistical variance refers to the average of the squared differences between the data sequence values ​​and the mean, reflecting the degree of data dispersion. The peak-to-peak ratio refers to the ratio of the signal's peak-to-peak value to the mean, characterizing the signal's dynamic range. The formula for statistical variance is: ,in, is the amplitude difference of the nth sampling point, is the mean of the amplitude difference sequence. is the total number of sampling points in the amplitude difference sequence. The peak-to-peak ratio formula is: ,in, is the maximum value of the amplitude difference sequence, is the minimum value of the amplitude difference sequence.

[0159] The following is a specific example: Taking a high-voltage cable monitoring as an example, in an online monitoring scenario, after real-time signal acquisition, the amplitude difference sequence of the three frequency bands of 1.2-1.8GHz is calculated to have a mean of 0.05V, 0.06V and 0.04V, and a variance of 0.002, 0.003 and 0.001. Combined with the energy proportions of 15%, 18% and 12%, the 9-dimensional time-frequency domain feature vectors of 0.05, 0.002, 15, 0.06, 0.003, 18, 0.04, 0.001 and 12 are generated in the order of the frequency bands. A lightweight convolutional neural network model enables real-time determination of discharge types. For offline monitoring scenarios, the variances of each frequency band are calculated for the same batch of historical data, with values ​​of 0.002, 0.003, and 0.001, peak-to-peak ratios of 3.0, 2.8, and 3.2, and energy proportions of 15%, 18%, and 12%. These are then fused to generate 9-dimensional feature vectors of 0.002, 3.0, 15, 0.003, 2.8, 18, 0.001, 3.2, and 12, which are then input into a support vector machine model for training and classification boundary optimization, thereby improving the generalization and accuracy of pattern recognition.

[0160] By executing steps c1 to c2, the embodiment of the present application realizes real-time classification through lightweight feature splicing in online scenarios, and introduces deep statistics such as peak-to-peak ratio in offline scenarios to enhance feature characterization capabilities, taking into account both real-time performance and analysis depth, adapting to different operation and maintenance requirements, and improving the scenario adaptability and model generalization of partial discharge pattern classification.

[0161] Figure 2This is a structural diagram of a high-voltage cable multi-parameter intelligent diagnostic system provided in an embodiment of the present application, such as Figure 2 As shown, the system includes:

[0162] The acquisition module 21 is used to deploy corresponding ultra-high frequency sensor arrays at the joints and terminals of the high-voltage cable to obtain partial discharge signals in the insulation layer of the high-voltage cable. The operating frequency band of the ultra-high frequency sensor array matches the dielectric spectrum characteristics of the high-voltage cable insulation material.

[0163] The decomposition module 22 is used to filter out power frequency and low-frequency interference from the partial discharge signal in the insulation layer of the high-voltage cable through a Butterworth high-pass filter, decompose the partial discharge signal after interference filtering using a multi-scale wavelet, and extract the time domain amplitude change rate and energy distribution parameters of each frequency band to generate a multi-dimensional time-frequency domain feature vector.

[0164] The calibration module 23 is used to calibrate the multidimensional time-frequency domain feature vector through the optical fiber synchronous transmission protocol, and use a local linear embedding algorithm to perform nonlinear manifold mapping on the calibrated multidimensional time-frequency domain feature vector to generate a low-dimensional feature vector adapted to the dielectric spectrum characteristics.

[0165] The matching module 24 is used to input the low-dimensional feature vector into a pre-built deep learning architecture, output a recognition result including a classification probability distribution, match the partial discharge type with the highest similarity to the recognition result from the historical discharge pattern database, and determine the insulation degradation level based on the final loss threshold interval in the dielectric spectrum characteristics, thereby generating a comprehensive diagnostic report including the partial discharge type and insulation degradation level.

[0166] Figure 2 The multi-parameter intelligent diagnosis system for high-voltage cables can be used to Figure 1 The implementation principles and technical effects of the multi-parameter intelligent diagnostic method for high-voltage cables described in the illustrated embodiment are not further elaborated. The specific manner in which each module and unit performs operations in the multi-parameter intelligent diagnostic system for high-voltage cables in the aforementioned embodiment has been described in detail in the relevant embodiments of the method and will not be further elaborated here.

[0167] In one possible design, Figure 2 A high voltage cable multi-parameter intelligent diagnosis system of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32 .

[0168] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0169] The processing component 32 is used to perform: deploying corresponding ultra-high frequency sensor arrays at the joints and terminal positions of the high-voltage cable respectively to obtain the local discharge signal in the insulation layer of the high-voltage cable, and matching the working frequency band of the ultra-high frequency sensor array with the dielectric spectrum characteristics of the high-voltage cable insulation material; filtering the local discharge signal in the insulation layer of the high-voltage cable through a Butterworth high-pass filter to remove the power frequency and low-frequency interference, using a multi-scale wavelet to decompose the local discharge signal after the interference is filtered out, and extracting the time domain amplitude change rate and energy distribution parameters of each frequency band to generate a multi-dimensional time-frequency domain feature vector; and transmitting the local discharge signal through the optical fiber synchronous transmission protocol. The multidimensional time-frequency domain feature vector is calibrated, and nonlinear manifold mapping is performed on the calibrated multidimensional time-frequency domain feature vector using a locally linear embedding algorithm to generate a low-dimensional feature vector that is adapted to the dielectric spectrum characteristics; the low-dimensional feature vector is input into a pre-built deep learning architecture, and an identification result including a classification probability distribution is output. The local discharge type with the highest similarity to the identification result is matched from a historical discharge pattern database, and the insulation degradation level is determined in combination with the final loss threshold interval in the dielectric spectrum characteristics, and a comprehensive diagnostic report including the local discharge type and the insulation degradation level is generated.

[0170] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0171] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as random access memory (RAM), static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0172] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0173] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0174] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0175] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0176] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 A multi-parameter intelligent diagnosis method for high-voltage cables according to the embodiment shown.

[0177] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0178] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0179] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0180] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-parameter intelligent diagnosis method for high-voltage cables, characterized in that: include: Deploy corresponding ultra-high frequency sensor arrays at the joints and terminals of the high-voltage cable to obtain partial discharge signals within the insulation layer of the high-voltage cable. The operating frequency band of the ultra-high frequency sensor array matches the dielectric spectrum characteristics of the high-voltage cable insulation material. The local discharge signal in the insulation layer of the high-voltage cable is filtered out of power frequency and low-frequency interference through a Butterworth high-pass filter, the local discharge signal after interference filtering is decomposed using a multi-scale wavelet, and the time domain amplitude change rate and energy distribution parameters of each frequency band are extracted to generate a multi-dimensional time-frequency domain feature vector; The multidimensional time-frequency domain feature vector is calibrated by using an optical fiber synchronous transmission protocol, and a nonlinear manifold mapping is performed on the calibrated multidimensional time-frequency domain feature vector using a local linear embedding algorithm to generate a low-dimensional feature vector adapted to the dielectric spectrum characteristics; Inputting the low-dimensional feature vector into a pre-built deep learning architecture, outputting a recognition result including a classification probability distribution, matching the partial discharge type with the highest similarity to the recognition result from a historical discharge pattern database, and determining the insulation degradation level based on the final loss threshold interval in the dielectric spectrum characteristics, thereby generating a comprehensive diagnostic report including the partial discharge type and insulation degradation level; The method uses a local linear embedding algorithm to perform nonlinear manifold mapping on the calibrated multidimensional time-frequency domain feature vector to generate a low-dimensional feature vector adapted to the dielectric spectrum characteristics, including: Determining a target local neighborhood radius corresponding to a multidimensional time-frequency domain feature vector based on a dielectric loss tangent value in the dielectric spectrum characteristics; calculating, within the local neighborhood radius, a spatial distance weight between multidimensional time-frequency domain feature vectors, wherein the spatial distance weight is inversely proportional to a dispersion parameter in the dielectric spectrum characteristic; Performing frequency band weighted correction on the spatial distance weight according to the frequency distribution in the dielectric spectrum characteristics to generate a corrected spatial distance weight; Constructing a local linear reconstruction relationship between multidimensional time-frequency domain feature vectors using the modified spatial distance weights, and extracting basis vectors in the local linear reconstruction relationship; Based on the absorption peak position in the dielectric spectrum characteristics, a low-dimensional feature vector corresponding to the absorption peak frequency band in the basis vector is retained.

2. The high-voltage cable multi-parameter intelligent diagnosis method according to claim 1, characterized in that: The determining of the target local neighborhood radius corresponding to the multi-dimensional time-frequency domain feature vector based on the dielectric loss tangent value in the dielectric spectrum characteristics includes: Extracting dielectric loss tangent values ​​at different frequency points in the dielectric spectrum characteristics based on the spectrum characteristics of the partial discharge signal after interference is filtered out, so as to construct a dielectric loss tangent value sequence; According to the dielectric loss tangent value sequence, an initial local neighborhood radius is obtained using a mapping table of the dielectric loss tangent value and the local neighborhood radius; The initial local neighborhood radius is compensated and corrected according to the thickness of the high-voltage cable insulation layer to generate a target local neighborhood radius, which is used to limit the spatial neighborhood boundary of the multidimensional time-frequency domain feature vector.

3. The high-voltage cable multi-parameter intelligent diagnosis method according to claim 1, characterized in that: The determining of the insulation degradation level based on the final loss threshold interval in the dielectric spectrum characteristics includes: Calculating the cumulative loss factor of the dielectric loss tangent value within the absorption peak frequency band; Constructing an intermediate loss threshold interval based on the exponential decay relationship between the cumulative loss factor and the thickness of the high-voltage cable insulation layer, wherein the upper limit of the intermediate loss threshold interval is determined by the frequency slope of the absorption peak frequency band; Performing frequency band sensitivity correction on the intermediate loss threshold interval based on the frequency dispersion parameter to generate a final loss threshold interval; Mapping the energy distribution of the time-frequency domain feature vector corresponding to the partial discharge type to the final loss threshold interval to calculate the deviation of the energy of each frequency band from the corresponding final loss threshold interval; The insulation degradation level is determined according to the cumulative proportion of the deviation within the absorption peak frequency band and the mapping relationship between the cumulative proportion and the insulation degradation level.

4. The high-voltage cable multi-parameter intelligent diagnosis method according to claim 3 is characterized in that: The step of constructing an intermediate loss threshold interval based on the exponential decay relationship between the cumulative loss factor and the thickness of the high-voltage cable insulation layer includes: Extracting the maximum and minimum values ​​of the cumulative loss factor within the absorption peak frequency band, and calculating the logarithmic attenuation ratios of the maximum and minimum values ​​to the thickness of the high-voltage cable insulation layer; Based on the logarithmic attenuation ratio and the frequency slope, constructing an exponential attenuation coefficient dynamic adjustment model; Calculating an exponential attenuation coefficient corresponding to the thickness of the high-voltage cable insulation layer using the exponential attenuation coefficient dynamic adjustment model; generating an initial loss threshold interval according to the exponential decay coefficient and the cumulative loss factor; Based on the confidence interval of the thickness of the high-voltage cable insulation layer, error compensation is performed on the initial loss threshold interval to generate an intermediate loss threshold interval.

5. The high-voltage cable multi-parameter intelligent diagnosis method according to claim 1, characterized in that: The method comprises filtering the power frequency and low-frequency interference of the local discharge signal in the insulation layer of the high-voltage cable through a Butterworth high-pass filter, decomposing the local discharge signal after the interference is filtered out using a multi-scale wavelet, extracting the time domain amplitude change rate and energy distribution parameters of each frequency band, and generating a multi-dimensional time-frequency domain feature vector, including: Based on the lower limit of the sensor's operating frequency band and the frequency value corresponding to the minimum dielectric constant in the dielectric spectrum characteristics, the cutoff frequency threshold of the Butterworth high-pass filter is set to filter out power frequency and low-frequency interference; Determining the number of levels of a multi-scale wavelet according to the number and intervals of absorption peak frequency bands in the dielectric spectrum characteristics, and using the multi-scale wavelet to perform hierarchical decomposition on the partial discharge signal after filtering out interference to obtain wavelet sub-band signals of each frequency band; For the wavelet subband signals of each frequency band, extracting the time domain amplitude change rate of each frequency band to form an amplitude difference sequence; Based on the attenuation gradient of the dielectric loss tangent value in the dielectric spectrum characteristics in the corresponding frequency band, performing frequency band weight correction on the time domain energy integral value of the wavelet subband signal of each frequency band to generate energy distribution parameters; The amplitude difference sequence and the energy distribution parameters are processed in a manner corresponding to the scene to form a multi-dimensional time-frequency domain feature vector.

6. The high-voltage cable multi-parameter intelligent diagnosis method according to claim 5, characterized in that: The processing of the amplitude difference sequence and the energy distribution parameter in a manner corresponding to the scene to form a multi-dimensional time-frequency domain feature vector includes: In an online monitoring scenario, the amplitude difference sequence and the energy distribution parameter are concatenated in frequency band order to form a multidimensional time-frequency domain feature vector; In an offline monitoring scenario, the statistical variance and peak-to-peak ratio of the amplitude difference sequence are calculated, and the statistical variance, the peak-to-peak ratio and the energy distribution parameter are concatenated in frequency band order to form a multidimensional time-frequency domain feature vector.

7. A high-voltage cable multi-parameter intelligent diagnostic system, characterized in that: include: An acquisition module is configured to deploy corresponding ultra-high frequency sensor arrays at the joints and terminals of the high-voltage cable to acquire partial discharge signals within the insulation layer of the high-voltage cable. The operating frequency band of the ultra-high frequency sensor array matches the dielectric spectrum characteristics of the high-voltage cable insulation material. a decomposition module for filtering out power frequency and low-frequency interference from the partial discharge signal in the insulation layer of the high-voltage cable through a Butterworth high-pass filter, decomposing the partial discharge signal after interference removal using a multi-scale wavelet, and extracting the time-domain amplitude change rate and energy distribution parameters of each frequency band to generate a multi-dimensional time-frequency domain feature vector; a calibration module for calibrating the multidimensional time-frequency domain feature vectors through an optical fiber synchronous transmission protocol, and performing nonlinear manifold mapping on the calibrated multidimensional time-frequency domain feature vectors using a locally linear embedding algorithm to generate low-dimensional feature vectors adapted to the dielectric spectrum characteristics; a matching module, configured to input the low-dimensional feature vector into a pre-built deep learning architecture, output a recognition result including a classification probability distribution, match the partial discharge type with the highest similarity to the recognition result from a historical discharge pattern database, determine the insulation degradation level based on the final loss threshold interval in the dielectric spectrum characteristics, and generate a comprehensive diagnostic report including the partial discharge type and insulation degradation level; The method uses a local linear embedding algorithm to perform nonlinear manifold mapping on the calibrated multidimensional time-frequency domain feature vector to generate a low-dimensional feature vector adapted to the dielectric spectrum characteristics, including: Determining a target local neighborhood radius corresponding to a multidimensional time-frequency domain feature vector based on a dielectric loss tangent value in the dielectric spectrum characteristics; calculating, within the local neighborhood radius, a spatial distance weight between multidimensional time-frequency domain feature vectors, wherein the spatial distance weight is inversely proportional to a dispersion parameter in the dielectric spectrum characteristic; Performing frequency band weighted correction on the spatial distance weight according to the frequency distribution in the dielectric spectrum characteristics to generate a corrected spatial distance weight; Constructing a local linear reconstruction relationship between multidimensional time-frequency domain feature vectors using the modified spatial distance weights, and extracting basis vectors in the local linear reconstruction relationship; Based on the absorption peak position in the dielectric spectrum characteristics, a low-dimensional feature vector corresponding to the absorption peak frequency band in the basis vector is retained.

8. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a high-voltage cable multi-parameter intelligent diagnosis method as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, a high-voltage cable multi-parameter intelligent diagnosis method according to any one of claims 1 to 6 is implemented.

Images