A method and system for detecting a high voltage cable insulation fault

Abstract

The application discloses a kind of detection methods and systems of high-voltage cable insulation fault, the method includes obtaining the leakage current corresponding to target high-voltage cable under different high-voltage excitation parameter combinations respectively, to carry out the calculation of equivalent capacitance and dielectric loss factor;According to voltage grade, voltage excitation frequency, equivalent capacitance and dielectric loss factor, the two-dimensional high-voltage dielectric spectrum of target high-voltage cable is constructed;The node attribute graph and adjacency matrix corresponding to two-dimensional high-voltage dielectric spectrum are obtained, to input the adjacency matrix and node attribute graph into the pre-trained high-voltage cable insulation fault detection model, and output the detection result corresponding to the insulation fault type and service life of target high-voltage cable;According to detection result, the fault operation instruction corresponding to target high-voltage cable is generated, and based on fault operation instruction, control preset operation and maintenance equipment executes maintenance operation to target high-voltage cable, to improve the stability of power system operation.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring technology, specifically to a method and system for detecting insulation faults in high-voltage cables. Background Technology

[0002] As power systems evolve towards higher voltage and more intelligent operation, the operating environment of high-voltage cables is becoming increasingly complex, making their insulation performance crucial for the safe and stable operation of the power system. Issues such as aging, moisture absorption, and dielectric breakdown in high-voltage cable insulation often lead to major power accidents. Therefore, timely monitoring of potential insulation faults in high-voltage cables, enabling maintenance personnel to control these faults based on the monitoring results, has become a vital technical feature for ensuring the safe operation of the power system.

[0003] In recent years, frequency domain dielectric spectroscopy, as a non-invasive and highly sensitive method for monitoring insulation status, has been widely used in low-voltage or medium-voltage cables. By analyzing the changing trends of the cable's dielectric parameters at different frequencies, it can reveal the insulation status inside low- and medium-voltage cables. However, due to the more complex dielectric response in high-voltage cables and the higher measurement requirements, traditional empirical analysis methods for analyzing frequency domain dielectric spectra have significant limitations in processing the high-dimensional and highly nonlinear spectral data of high-voltage cables. They cannot accurately assess the insulation status of high-voltage cables. Furthermore, existing technologies can only identify insulation faults but cannot identify the types of insulation faults. This results in the inability to promptly and accurately maintain the target high-voltage cable when a fault occurs, affecting the stable operation of the power system. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention discloses a method and system for detecting insulation faults in high-voltage cables, which can be used to improve the stability of power system operation.

[0005] To achieve the above objectives, this invention discloses a method for detecting insulation faults in high-voltage cables, comprising:

[0006] The target high-voltage cable is subjected to multi-stage high-voltage excitation to obtain the leakage current of the target high-voltage cable under different combinations of high-voltage excitation parameters; wherein, the high-voltage excitation parameter combination includes voltage level and voltage excitation frequency;

[0007] Based on the leakage current, obtain the equivalent capacitance and dielectric loss factor of the target high-voltage cable under each of the high-voltage excitation parameter combinations;

[0008] Using the voltage level and the voltage excitation frequency as coordinate axes, and the equivalent capacitance and the dielectric loss factor as coordinate values, a two-dimensional high-voltage dielectric spectrum corresponding to the target high-voltage cable is constructed.

[0009] The node attribute graph and adjacency matrix corresponding to the two-dimensional high voltage dielectric spectrum are obtained according to the preset node construction strategy and edge construction strategy.

[0010] The adjacency matrix and the node attribute graph are input into a pre-trained high-voltage cable insulation fault detection model to output the detection results of the insulation fault type and service life of the target high-voltage cable.

[0011] Based on the detection results, a fault maintenance operation instruction corresponding to the target high-voltage cable is generated, and based on the fault maintenance operation instruction, a preset maintenance equipment is controlled to perform maintenance operations on the target high-voltage cable.

[0012] This invention discloses a method for detecting insulation faults in high-voltage cables. By combining multi-dimensional excitation parameter combinations with a model, it achieves accurate assessment and proactive maintenance of the insulation state of high-voltage cables. First, a multi-level high-voltage excitation parameter combination based on voltage level and voltage application frequency is used to obtain leakage current, overcoming the limitations of single-frequency or voltage excitation and enabling the stimulation of dielectric response characteristics of different insulation defects. Second, a two-dimensional dielectric spectrum is constructed using equivalent capacitance and dielectric loss factor as dual parameters, mapping high-dimensional nonlinear data into a structured spectrum, providing a suitable input format for subsequent model processing. Furthermore, the graph structure data construction based on node attributes and adjacency matrices can characterize the correlation of dielectric parameters under different excitation parameters, capturing the distribution pattern of insulation state in the voltage-frequency space. The pre-trained model integrates local and global features through a graph attention mechanism, solving the problem of traditional methods being unable to handle complex spectrum correlations. It also achieves accurate output of fault type classification and service life regression. Finally, maintenance instructions matched to the output results are used to perform maintenance on the target high-voltage cable, improving the stability of power system operation.

[0013] As a preferred example, the step of performing multi-stage high-voltage excitation on the target high-voltage cable to obtain the leakage current of the target high-voltage cable under different combinations of high-voltage excitation parameters includes:

[0014] Multiple sets of high-voltage excitation parameter combinations are generated based on multiple preset voltage excitation frequencies and multiple preset voltage levels; wherein, different high-voltage excitation parameter combinations have different voltage excitation frequencies or different voltage levels.

[0015] For any set of the high-voltage excitation parameter combinations:

[0016] An AC excitation signal corresponding to the high-voltage excitation parameter combination is generated, and the AC excitation signal is applied to both ends of the target high-voltage cable to measure the leakage current of the target high-voltage cable under the high-voltage excitation parameter combination.

[0017] The above scheme systematically solves the problem of a single excitation mode in high-voltage cable insulation condition detection by establishing a multi-dimensional excitation parameter space. First, by pre-setting multiple combinations of voltage excitation frequencies and voltage levels, a high-voltage excitation parameter combination covering different operating conditions is formed. This orthogonal parameter combination design breaks through the traditional single-frequency or fixed-voltage excitation mode, enabling the stimulation of the cable's dielectric response characteristics under different electrical stresses. Second, a corresponding AC excitation signal is generated for each set of parameters. By applying AC signals of specific frequencies and voltage levels at both ends of the cable, accurate measurement of leakage current under different electrical conditions is achieved. The dual variation of voltage level and frequency effectively simulates the complex electrical environment experienced by the cable in actual operation, providing multi-dimensional, high-resolution raw data support for the subsequent construction of high-voltage dielectric spectra. This parameterized excitation method avoids data redundancy caused by random excitation and ensures the detectability of different insulation defect characteristics under specific frequency-voltage combinations.

[0018] As a preferred example, obtaining the equivalent capacitance and dielectric loss factor of the target high-voltage cable under each combination of high-voltage excitation parameters based on the leakage current includes:

[0019] For any set of the high-voltage excitation parameter combinations:

[0020] Based on the voltage corresponding to the voltage level in the high-voltage excitation parameter combination and the leakage current of the target high-voltage cable under the high-voltage excitation parameter combination, the insulation impedance of the target high-voltage cable under the high-voltage excitation parameter combination is obtained.

[0021] The angular frequency of the AC excitation signal is determined based on the voltage excitation frequency in the high-voltage excitation parameter combination.

[0022] Based on the insulation impedance, the angular frequency, and the voltage excitation frequency, the complex capacitance of the target high-voltage cable under the given combination of high-voltage excitation parameters is obtained.

[0023] Separate the real part and imaginary part of the equivalent capacitance of the target high-voltage cable under the given high-voltage excitation parameter combination from the complex capacitance;

[0024] Based on the real part and the imaginary part of the equivalent capacitance, the dielectric loss factor of the target high-voltage cable under the given high-voltage excitation parameter combination is obtained.

[0025] The above scheme solves the problem of traditional methods being unable to handle the complex dielectric response of high-voltage cables by establishing a direct correlation between voltage level, leakage current, and insulation impedance. Specifically, the insulation impedance is calculated based on the voltage value corresponding to the voltage level and the measured leakage current, which can accurately reflect the insulation performance under different excitation conditions; the angular frequency is derived by the voltage excitation frequency, providing frequency domain parameter support for the calculation of complex capacitance; the complex capacitance is calculated by combining insulation impedance, angular frequency, and voltage excitation frequency, realizing the mathematical modeling of the polarization characteristics of the cable dielectric; the complex capacitance is decomposed into the real and imaginary parts of the equivalent capacitance, which respectively characterize the dielectric energy storage and loss characteristics, and finally the dielectric loss factor is obtained by the ratio of the two, thus fully constructing a multi-dimensional evaluation system for the insulation state of high-voltage cables.

[0026] As a preferred example, obtaining the equivalent capacitance and dielectric loss factor of the target high-voltage cable under each combination of high-voltage excitation parameters based on the leakage current includes:

[0027] Obtain a first ratio of the voltage divided by the leakage current, use the first ratio as the insulation resistance, and obtain a first reciprocal of the insulation resistance;

[0028] Obtain the frequency product of the preset imaginary unit and the angular frequency, and obtain the second reciprocal of the frequency product, so that the product of the first reciprocal and the second reciprocal can be used as the complex capacitance;

[0029] Obtain a second ratio of the imaginary part of the equivalent capacitance to the real part of the equivalent capacitance, and use the second ratio as the dielectric loss factor of the target high-voltage cable under the high-voltage excitation parameter combination.

[0030] The above scheme simplifies the calculation of equivalent capacitance and dielectric loss factor by establishing a direct mathematical relationship between voltage, leakage current, and frequency parameters. First, the insulation impedance is defined using the ratio of voltage to leakage current, and converted into admittance characteristics through reciprocal operations, providing a foundation for subsequent complex capacitance calculations. Second, the admittance characteristics are integrated with the frequency parameters by combining the reciprocal of the product of the voltage excitation frequency and the angular frequency, forming a mathematical expression for the complex capacitance, avoiding the complex equivalent circuit modeling required in traditional methods. Finally, the dielectric loss factor is directly characterized by the ratio of the imaginary to the real part of the equivalent capacitance, and the loss components are separated using the physical properties of the complex domain, solving the error accumulation problem caused by traditional methods relying on empirical formulas or approximate calculations. This series of mathematical steps achieves an efficient and linear conversion from raw measurement data to key dielectric parameters, providing accurate input data for the subsequent construction of a two-dimensional dielectric spectrum.

[0031] As a preferred example, the construction of a two-dimensional high-voltage dielectric spectrum corresponding to the target high-voltage cable, using the voltage level and the voltage excitation frequency as coordinate axes and the equivalent capacitance and the dielectric loss factor as coordinate values, includes:

[0032] For any set of the high-voltage excitation parameter combinations:

[0033] The real part of the equivalent capacitance, the imaginary part of the equivalent capacitance, and the dielectric loss factor corresponding to the high-voltage excitation parameter combination are normalized to obtain the normalized real part of the equivalent capacitance, the normalized imaginary part of the equivalent capacitance, and the normalized dielectric loss factor corresponding to the high-voltage excitation parameter combination.

[0034] According to the preset logarithmic step size and sorting rules from low to high, the multiple voltage excitation frequencies in the multiple high voltage excitation parameter combinations are arranged to obtain the frequency coordinate axis;

[0035] According to a preset ascending order of voltage levels, the voltage levels in the multiple combinations of high-voltage excitation parameters are arranged to obtain a voltage coordinate axis.

[0036] A two-dimensional matrix is ​​constructed based on the frequency coordinate axis and the voltage coordinate axis. The real part of the normalized equivalent capacitance, the imaginary part of the normalized equivalent capacitance, and the normalized dielectric loss factor corresponding to each combination of high-voltage excitation parameters are embedded into the two-dimensional matrix to obtain the two-dimensional high-voltage dielectric spectrum corresponding to the target high-voltage cable.

[0037] The above scheme eliminates the differences between parameters of different dimensions through data normalization, uses logarithmic step-size sorting of the frequency coordinate axes to capture subtle changes in the high-frequency band, constructs a regular distribution of the voltage coordinate axes through ascending voltage order, and finally maps multidimensional parameters to a two-dimensional space to form a structured spectrum. The embedding operations of the real and imaginary parts of the normalized equivalent capacitance and the dielectric loss factor achieve effective fusion of multidimensional insulation characteristics. The preset logarithmic step-size sorting strategy can adapt to the nonlinear response characteristics of high-voltage cables, and the ascending voltage order rule strengthens the correlation between voltage gradient changes and dielectric loss. By constructing a two-dimensional matrix with spatial topological relationships, an analytical high-dimensional feature representation carrier is provided for the subsequent graph neural network model, solving the core problem that traditional methods cannot handle complex dielectric spectrum data of high-voltage cables.

[0038] As a preferred example, the step of obtaining the node attribute graph and adjacency matrix corresponding to the two-dimensional high-voltage dielectric spectrum according to the preset node construction strategy and edge construction strategy includes:

[0039] Based on a preset node extraction strategy, multiple nodes are extracted from the two-dimensional high voltage dielectric spectrum to form a node set; wherein each node corresponds to a combination point of voltage excitation frequency and voltage level.

[0040] According to a preset connection strategy, physical connection relationships between nodes in the node set are generated to form an edge set; wherein, the connection strategy includes a frequency connection strategy, a voltage connection strategy, and a diagonal connection strategy; wherein, the frequency connection strategy connects nodes that are adjacent to each other on the frequency coordinate axis at the same voltage level; the voltage connection strategy connects nodes that are adjacent to each other on the voltage coordinate axis at the same voltage excitation frequency; the diagonal connection strategy connects nodes that are adjacent to each other on both the frequency coordinate axis and the voltage coordinate axis.

[0041] The normalized real part of the equivalent capacitance, the normalized imaginary part of the equivalent capacitance, and the normalized dielectric loss factor corresponding to each node are obtained from the two-dimensional high voltage dielectric spectrum as the feature vector of the node.

[0042] Generate a node attribute map corresponding to the two-dimensional high voltage dielectric spectrum based on the feature vector, the edge set, and the feature vector;

[0043] The adjacency matrix corresponding to the two-dimensional high-voltage dielectric spectrum is generated based on the topology of the node attribute graph.

[0044] The above scheme solves the problem that traditional methods cannot effectively express the spatial correlation of high-voltage dielectric spectra by constructing graph structure data with topological relationships. Specifically, a node set is constructed based on voltage-frequency combination points, transforming each test condition parameter into a graph node to ensure coverage of all test conditions. An edge set is established through three strategies: frequency connection, voltage connection, and diagonal connection. This preserves the continuity of frequency changes under the same voltage, captures the correlation of voltage changes under the same frequency, and achieves cross-dimensional spatial relationship modeling through diagonal connections. The normalized real and imaginary parts of the equivalent capacitance and the dielectric loss factor are used as multi-dimensional feature vectors to fully characterize the dielectric properties of each test point. The finally generated node attribute graph and adjacency matrix transform high-dimensional nonlinear spectrum data into structured graph data, providing input data for subsequent graph neural network models that can characterize both local node characteristics and reflect global topological relationships.

[0045] As a preferred example, the step of inputting the adjacency matrix and the node attribute graph into a pre-trained high-voltage cable insulation fault detection model to output the detection results of the insulation fault type and service life of the target high-voltage cable includes:

[0046] The adjacency matrix and the node attribute graph are synchronously input into the pre-trained high-voltage cable insulation fault detection model so as to extract the embedding features corresponding to each node in the node attribute graph through the high-voltage cable insulation fault detection model.

[0047] For any node structure relationship preset in the high-voltage cable insulation fault detection model:

[0048] The multiple embedded features obtained are averaged according to the node structure relationship to obtain the global feature vector corresponding to the node structure relationship;

[0049] The multiple global feature vectors are adaptively weighted and fused by an attention mechanism pre-set in the high-voltage cable insulation fault detection model to obtain a fused global feature vector.

[0050] The fused global feature vector is regressed and classified according to the regression classification function preset in the high-voltage cable insulation fault detection model, and the insulation status assessment result and equivalent service life of the target high-voltage cable are obtained and output.

[0051] The above scheme addresses the limitation of traditional methods in handling high-dimensional nonlinear features by constructing a graph neural network model to process two-dimensional high-voltage dielectric spectrum data. Synchronously inputting the adjacency matrix and node attribute graph into the model preserves the topological correlation characteristics of cable insulation status in the voltage-frequency space. When extracting the embedded features of each node, combining the adjacency matrix to identify neighboring node information enhances the spatial correlation of local features. A mean operation is performed on different node structural relationships to generate global feature vectors, effectively integrating insulation status information at different structural levels. An attention mechanism is used to adaptively weight and fuse multiple sets of global features, overcoming the shortcomings of traditional weighted averaging methods in distinguishing the contribution of different structural relationships. Finally, by jointly outputting fault type and service life through a regression classification function, a multi-dimensional comprehensive assessment of cable insulation status is achieved, improving detection accuracy.

[0052] As a preferred example, the step of synchronously inputting the adjacency matrix and the node attribute graph into a pre-trained high-voltage cable insulation fault detection model, so as to extract the embedding features corresponding to each node in the node attribute graph through the high-voltage cable insulation fault detection model, includes:

[0053] For any node in the node attribute graph:

[0054] Based on the adjacency matrix, the high-voltage cable insulation fault detection model identifies multiple adjacent nodes corresponding to the node, and extracts the first feature information of the node and the second feature information of each of the adjacent nodes.

[0055] The weights corresponding to each of the second feature information are assigned by the attention mechanism preset in the high-voltage cable insulation fault detection model.

[0056] Based on the weights, the first feature information and multiple second feature information are weighted and fused using the high-voltage cable insulation fault detection model to obtain the embedded feature corresponding to the node.

[0057] The above scheme solves the problem of simultaneously considering local features and global correlations in high-voltage cable dielectric spectrum data by constructing a dynamic correlation feature fusion mechanism between nodes. Specifically, it identifies adjacent nodes based on the adjacency matrix, breaking through the limitation of traditional methods that only focus on single node features. By extracting the second feature information of adjacent nodes, it establishes the physical connection relationship and data correlation between nodes. An attention mechanism is used to allocate weights, which can adaptively adjust the feature fusion ratio according to the actual correlation strength between nodes, avoiding empirical errors from manually setting weights. Finally, a weighted fusion method integrates the first feature information of the node itself with the second feature information of adjacent nodes, so that the generated embedded features contain both local node attributes and spatial correlation information in the topological structure. This feature extraction method based on a graph structure attention mechanism significantly improves the model's ability to represent complex dielectric spectrum features, providing high-dimensional feature support for subsequent accurate identification of insulation fault types and assessment of service life.

[0058] On the other hand, the present invention discloses a high-voltage cable insulation fault detection system, including a multi-level high-voltage excitation module, a fault medium extraction module, a spectrum mapping module, a node attribute extraction module, an insulation fault detection module, and a cable fault operation and maintenance module;

[0059] The multi-stage high-voltage excitation module is used to perform multi-stage high-voltage excitation on the target high-voltage cable to obtain the leakage current of the target high-voltage cable under different combinations of high-voltage excitation parameters; wherein, the high-voltage excitation parameter combination includes voltage level and voltage excitation frequency;

[0060] The fault medium extraction module is used to obtain the equivalent capacitance and dielectric loss factor of the target high-voltage cable under each combination of high-voltage excitation parameters based on the leakage current.

[0061] The spectrum mapping module is used to construct a two-dimensional high-voltage dielectric spectrum corresponding to the target high-voltage cable, with the voltage level and the voltage excitation frequency as coordinate axes and the equivalent capacitance and the dielectric loss factor as coordinate values.

[0062] The node attribute extraction module is used to obtain the node attribute graph and adjacency matrix corresponding to the two-dimensional high voltage dielectric spectrum according to the preset node construction strategy and edge construction strategy.

[0063] The insulation fault detection module is used to input the adjacency matrix and the node attribute graph into the pre-trained high-voltage cable insulation fault detection model, so as to output the detection results of the insulation fault type and service life of the target high-voltage cable.

[0064] The cable fault maintenance module is used to generate fault maintenance operation instructions corresponding to the target high-voltage cable based on the detection results, and control the preset maintenance equipment to perform maintenance operations on the target high-voltage cable based on the fault maintenance operation instructions.

[0065] This invention discloses a high-voltage cable insulation fault detection system that achieves accurate assessment and proactive maintenance of high-voltage cable insulation status through a combination of multi-dimensional excitation parameters and a model. First, it employs a multi-level high-voltage excitation parameter combination based on voltage level and applied voltage frequency to obtain leakage current, overcoming the limitations of single-frequency or voltage excitation and enabling the stimulation of dielectric response characteristics of different insulation defects. Second, it constructs a two-dimensional dielectric spectrum using equivalent capacitance and dielectric loss factor as dual parameters, mapping high-dimensional nonlinear data into a structured spectrum, providing a suitable input format for subsequent model processing. Furthermore, the graph structure data construction based on node attributes and adjacency matrices characterizes the correlation of dielectric parameters under different excitation parameters, capturing the distribution pattern of insulation status in the voltage-frequency space. The pre-trained model integrates local and global features through a graph attention mechanism, solving the problem of traditional methods being unable to handle complex spectrum correlations, while simultaneously achieving accurate output of fault type classification and service life regression. Finally, based on the output results, maintenance instructions are matched to perform maintenance on the target high-voltage cable, improving the stability of power system operation.

[0066] As a preferred example, the multi-stage high-voltage excitation module includes a parameter combination unit and an excitation unit;

[0067] The parameter combination unit is used to generate multiple sets of high-voltage excitation parameter combinations based on multiple preset voltage excitation frequencies and multiple preset voltage levels; wherein, different high-voltage excitation parameter combinations have different voltage excitation frequencies or different voltage levels;

[0068] The excitation unit is used to generate an AC excitation signal corresponding to any set of high-voltage excitation parameter combinations, and apply the AC excitation signal to both ends of the target high-voltage cable to measure the leakage current of the target high-voltage cable under the high-voltage excitation parameter combination.

[0069] The above scheme systematically solves the problem of a single excitation mode in high-voltage cable insulation condition detection by establishing a multi-dimensional excitation parameter space. First, by pre-setting multiple combinations of voltage excitation frequencies and voltage levels, a high-voltage excitation parameter combination covering different operating conditions is formed. This orthogonal parameter combination design breaks through the traditional single-frequency or fixed-voltage excitation mode, enabling the stimulation of the cable's dielectric response characteristics under different electrical stresses. Second, a corresponding AC excitation signal is generated for each set of parameters. By applying AC signals of specific frequencies and voltage levels at both ends of the cable, accurate measurement of leakage current under different electrical conditions is achieved. The dual variation of voltage level and frequency effectively simulates the complex electrical environment experienced by the cable in actual operation, providing multi-dimensional, high-resolution raw data support for the subsequent construction of high-voltage dielectric spectra. This parameterized excitation method avoids data redundancy caused by random excitation and ensures the detectability of different insulation defect characteristics under specific frequency-voltage combinations. Attached Figure Description

[0070] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0071] Figure 1 This is a schematic flowchart of a method for detecting insulation faults in high-voltage cables disclosed in an embodiment of the present invention;

[0072] Figure 2 This is a schematic diagram of the structure of a high-voltage cable insulation fault detection system disclosed in an embodiment of the present invention;

[0073] Figure 3 This is a flowchart illustrating a method for detecting insulation faults in high-voltage cables, as disclosed in another embodiment of the present invention. Detailed Implementation

[0074] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] Example 1

[0076] Reference Figure 1To address the limitations of existing technologies in timely and accurate detection of insulation faults in high-voltage cables, thereby improving the stability of power system operation, this embodiment discloses a method for detecting insulation faults in high-voltage cables, mainly comprising:

[0077] Step 101: Perform multi-stage high-voltage excitation on the target high-voltage cable to obtain the leakage current of the target high-voltage cable under different combinations of high-voltage excitation parameters; wherein, the combination of high-voltage excitation parameters includes voltage level and voltage excitation frequency.

[0078] In this embodiment, the main steps are as follows: generating multiple sets of high-voltage excitation parameter combinations based on multiple preset voltage excitation frequencies and multiple preset voltage levels; wherein, different high-voltage excitation parameter combinations have different voltage excitation frequencies or different voltage levels; for any set of high-voltage excitation parameter combinations: generating an AC excitation signal corresponding to the high-voltage excitation parameter combination, and applying the AC excitation signal to both ends of the target high-voltage cable to measure the leakage current of the target high-voltage cable under the high-voltage excitation parameter combination.

[0079] In this embodiment, the above steps systematically solve the problem of a single excitation mode in high-voltage cable insulation condition detection by establishing a multi-dimensional excitation parameter space. First, by pre-setting multiple combinations of voltage excitation frequencies and voltage levels, a high-voltage excitation parameter combination covering different operating conditions is formed. This orthogonal parameter combination design breaks through the traditional single-frequency or fixed-voltage excitation mode, enabling the excitation of the cable's dielectric response characteristics under different electrical stresses. Second, a corresponding AC excitation signal is generated for each set of parameters. By applying AC signals of specific frequencies and voltage levels to both ends of the cable, accurate measurement of leakage current under different electrical conditions is achieved. The dual variation of voltage level and frequency effectively simulates the complex electrical environment experienced by the cable in actual operation, providing multi-dimensional, high-resolution raw data support for the subsequent construction of high-voltage dielectric spectra. This parameterized excitation method avoids data redundancy caused by random excitation and ensures the detectability of different insulation defect characteristics under specific frequency-voltage combinations.

[0080] Step 102: Obtain the equivalent capacitance and dielectric loss factor of the target high-voltage cable under each combination of high-voltage excitation parameters based on the leakage current.

[0081] In this embodiment, the step mainly includes: for any set of high-voltage excitation parameter combinations: obtaining the insulation impedance of the target high-voltage cable under the high-voltage excitation parameter combination based on the voltage corresponding to the voltage level in the high-voltage excitation parameter combination and the leakage current of the target high-voltage cable under the high-voltage excitation parameter combination; determining the angular frequency of the AC excitation signal based on the voltage excitation frequency in the high-voltage excitation parameter combination; obtaining the complex capacitance of the target high-voltage cable under the high-voltage excitation parameter combination based on the insulation impedance, the angular frequency, and the voltage excitation frequency; separating the real part and the imaginary part of the equivalent capacitance of the target high-voltage cable under the high-voltage excitation parameter combination from the complex capacitance; and obtaining the dielectric loss factor of the target high-voltage cable under the high-voltage excitation parameter combination based on the real part and the imaginary part of the equivalent capacitance.

[0082] Specifically, the method involves obtaining a first ratio of the voltage divided by the leakage current, using this first ratio as the insulation impedance, and obtaining a first reciprocal of the insulation impedance; obtaining a preset imaginary unit multiplied by the frequency of the angular frequency, and obtaining a second reciprocal of the frequency product, using the product of the first reciprocal and the second reciprocal as the complex capacitance; and obtaining a second ratio of the imaginary part of the equivalent capacitance divided by the real part of the equivalent capacitance, using this second ratio as the dielectric loss factor of the target high-voltage cable under the high-voltage excitation parameter combination.

[0083] In this embodiment, the above steps establish a direct correlation between voltage level, leakage current, and insulation impedance, solving the problem that traditional methods cannot handle the complex dielectric response of high-voltage cables. Specifically, the insulation impedance is calculated based on the voltage value corresponding to the voltage level and the measured leakage current, which can accurately reflect the insulation performance under different excitation conditions; the angular frequency is derived by the voltage excitation frequency, providing frequency domain parameter support for the calculation of complex capacitance; the complex capacitance is calculated by combining insulation impedance, angular frequency, and voltage excitation frequency, realizing the mathematical modeling of the polarization characteristics of the cable dielectric; the complex capacitance is decomposed into the real and imaginary parts of the equivalent capacitance, which respectively characterize the dielectric energy storage and loss characteristics, and finally the dielectric loss factor is obtained by the ratio of the two, thus fully constructing a multi-dimensional evaluation system for the insulation state of high-voltage cables.

[0084] Step 103: Using the voltage level and the voltage excitation frequency as coordinate axes, and the equivalent capacitance and the dielectric loss factor as coordinate values, construct a two-dimensional high-voltage dielectric spectrum corresponding to the target high-voltage cable.

[0085] In this embodiment, this step mainly includes: for any set of the high-voltage excitation parameter combinations:

[0086] The real part, imaginary part, and dielectric loss factor of the equivalent capacitance corresponding to the high-voltage excitation parameter combination are normalized to obtain the normalized real part, imaginary part, and dielectric loss factor of the equivalent capacitance corresponding to the high-voltage excitation parameter combination. Multiple voltage excitation frequencies in the multiple high-voltage excitation parameter combinations are arranged according to a preset logarithmic step size and a sorting rule from low to high to obtain a frequency coordinate axis. Multiple voltage levels in the multiple high-voltage excitation parameter combinations are arranged according to a preset ascending order of voltage levels to obtain a voltage coordinate axis. A two-dimensional matrix is ​​constructed based on the frequency coordinate axis and the voltage coordinate axis, and the normalized real part, imaginary part, and dielectric loss factor of the equivalent capacitance corresponding to each high-voltage excitation parameter combination are embedded into the two-dimensional matrix to obtain the two-dimensional high-voltage dielectric spectrum corresponding to the target high-voltage cable.

[0087] In this embodiment, the above steps eliminate the differences between parameters of different dimensions through data normalization, employ logarithmic step-size sorting of the frequency coordinate axes to capture subtle changes in the high-frequency band, construct a regular distribution of the voltage coordinate axes through ascending voltage order, and finally map the multidimensional parameters to a two-dimensional space to form a structured spectrum. The embedding operations of the real and imaginary parts of the normalized equivalent capacitance and the dielectric loss factor effectively fuse multidimensional insulation characteristics. The preset logarithmic step-size sorting strategy can adapt to the nonlinear response characteristics of high-voltage cables, and the ascending voltage order rule strengthens the correlation between voltage gradient changes and dielectric loss. By constructing a two-dimensional matrix with spatial topological relationships, a resolvable high-dimensional feature expression carrier is provided for the subsequent graph neural network model, solving the core problem that traditional methods cannot handle complex dielectric spectrum data of high-voltage cables.

[0088] Step 104: Obtain the node attribute graph and adjacency matrix corresponding to the two-dimensional high voltage dielectric spectrum according to the preset node construction strategy and edge construction strategy.

[0089] In this embodiment, the steps mainly include: extracting multiple nodes from the two-dimensional high-voltage dielectric spectrum based on a preset node extraction strategy to form a node set; wherein each node corresponds to a combination point of voltage excitation frequency and voltage level; generating physical connection relationships between the nodes in the node set to form an edge set according to a preset connection strategy; wherein the connection strategy includes a frequency connection strategy, a voltage connection strategy, and a diagonal connection strategy; wherein the frequency connection strategy connects nodes at the same voltage level that are adjacent to each other on the frequency coordinate axis; the voltage connection strategy connects nodes at the same voltage excitation frequency that are adjacent to each other on the voltage coordinate axis; the diagonal connection strategy connects nodes that are adjacent to each other on both the frequency coordinate axis and the voltage coordinate axis; obtaining the normalized equivalent capacitance real part, the normalized equivalent capacitance imaginary part, and the normalized dielectric loss factor corresponding to each node from the two-dimensional high-voltage dielectric spectrum as the feature vector of the node; generating a node attribute graph corresponding to the two-dimensional high-voltage dielectric spectrum based on the feature vector, the edge set, and the feature vector; and generating an adjacency matrix corresponding to the two-dimensional high-voltage dielectric spectrum based on the topology of the node attribute graph.

[0090] In this embodiment, the above steps solve the problem that traditional methods cannot effectively express the spatial correlation of high-voltage dielectric spectra by constructing graph structure data with topological relationships. Specifically, a node set is constructed based on voltage-frequency combination points, and each test condition parameter is transformed into a graph node to ensure coverage of all test conditions. An edge set is established through three strategies: frequency connection, voltage connection, and diagonal connection. This preserves the continuity of frequency changes under the same voltage and captures the correlation of voltage changes under the same frequency. At the same time, diagonal connection realizes the modeling of cross-dimensional spatial relationships. The normalized real part, imaginary part of the equivalent capacitance, and dielectric loss factor are used as multi-dimensional feature vectors to fully characterize the dielectric properties of each test point. The finally generated node attribute graph and adjacency matrix transform the high-dimensional nonlinear spectrum data into structured graph data, providing input data for subsequent graph neural network models that can characterize both local node characteristics and reflect global topological relationships.

[0091] Step 105: Input the adjacency matrix and the node attribute graph into the pre-trained high-voltage cable insulation fault detection model to output the detection results of the insulation fault type and service life of the target high-voltage cable.

[0092] In this embodiment, the steps mainly include: synchronously inputting the adjacency matrix and the node attribute graph into a pre-trained high-voltage cable insulation fault detection model, so as to extract the embedded features corresponding to each node in the node attribute graph through the high-voltage cable insulation fault detection model; for any node structural relationship preset in the high-voltage cable insulation fault detection model: performing a mean operation on the multiple embedded features obtained according to the node structural relationship to obtain the global feature vector corresponding to the node structural relationship; adaptively weighting and fusing the multiple global feature vectors through the attention mechanism preset in the high-voltage cable insulation fault detection model to obtain a fused global feature vector; performing regression classification on the fused global feature vector according to the regression classification function preset in the high-voltage cable insulation fault detection model to obtain and output the insulation status assessment result and equivalent service life of the target high-voltage cable.

[0093] Specifically, for any node in the node attribute graph: based on the adjacency matrix, multiple adjacent nodes corresponding to the node are identified through the high-voltage cable insulation fault detection model, and the first feature information of the node and the second feature information of each of the adjacent nodes are extracted; weights corresponding to each of the second feature information are assigned through an attention mechanism preset in the high-voltage cable insulation fault detection model; based on the weights, the first feature information and multiple second feature information are weighted and fused through the high-voltage cable insulation fault detection model to obtain the embedded features corresponding to the node.

[0094] In this embodiment, the above steps utilize a graph neural network model to process two-dimensional high-voltage dielectric spectrum data, overcoming the limitation of traditional methods in handling high-dimensional nonlinear features. Synchronously inputting the adjacency matrix and node attribute graph into the model preserves the topological correlation characteristics of cable insulation status in the voltage-frequency space. When extracting the embedded features of each node, combining the adjacency matrix to identify neighboring node information enhances the spatial correlation of local features. A mean operation is performed on different node structural relationships to generate global feature vectors, effectively integrating insulation status information at different structural levels. An attention mechanism is employed to adaptively weight and fuse multiple sets of global features, overcoming the shortcomings of traditional weighted averaging methods in distinguishing the contribution of different structural relationships. Finally, a regression classification function is used to jointly output fault type and service life, achieving a multi-dimensional comprehensive evaluation of cable insulation status and improving detection accuracy.

[0095] Step 106: Generate a fault maintenance operation instruction corresponding to the target high-voltage cable based on the detection results, and control the preset maintenance equipment to perform maintenance operations on the target high-voltage cable based on the fault maintenance operation instruction.

[0096] On the other hand, refer to Figure 2This embodiment also discloses a high-voltage cable insulation fault detection system, including a multi-level high-voltage excitation module 201, a fault medium extraction module 202, a spectrum mapping module 203, a node attribute extraction module 204, an insulation fault detection module 205, and a cable fault maintenance module 206.

[0097] The multi-level high-voltage excitation module 201 is used to perform multi-level high-voltage excitation on the target high-voltage cable to obtain the leakage current of the target high-voltage cable under different combinations of high-voltage excitation parameters; wherein, the high-voltage excitation parameter combination includes voltage level and voltage excitation frequency.

[0098] The fault medium extraction module 202 is used to obtain the equivalent capacitance and dielectric loss factor of the target high-voltage cable under each combination of high-voltage excitation parameters based on the leakage current.

[0099] The spectrum mapping module 203 is used to construct a two-dimensional high-voltage dielectric spectrum corresponding to the target high-voltage cable, with the voltage level and the voltage excitation frequency as coordinate axes and the equivalent capacitance and the dielectric loss factor as coordinate values.

[0100] The node attribute extraction module 204 is used to obtain the node attribute graph and adjacency matrix corresponding to the two-dimensional high voltage dielectric spectrum according to the preset node construction strategy and edge construction strategy.

[0101] The insulation fault detection module 205 is used to input the adjacency matrix and the node attribute graph into the pre-trained high-voltage cable insulation fault detection model, so as to output the detection results of the insulation fault type and service life of the target high-voltage cable.

[0102] The cable fault maintenance module 206 is used to generate a fault maintenance operation instruction corresponding to the target high-voltage cable based on the detection result, and control the preset maintenance equipment to perform maintenance operations on the target high-voltage cable based on the fault maintenance operation instruction.

[0103] In this embodiment, the multi-stage high-voltage excitation module 201 includes a parameter combination unit and an excitation unit;

[0104] The parameter combination unit is used to generate multiple sets of high-voltage excitation parameter combinations based on multiple preset voltage excitation frequencies and multiple preset voltage levels; wherein, different high-voltage excitation parameter combinations have different voltage excitation frequencies or different voltage levels;

[0105] The excitation unit is used to generate an AC excitation signal corresponding to any set of high-voltage excitation parameter combinations, and apply the AC excitation signal to both ends of the target high-voltage cable to measure the leakage current of the target high-voltage cable under the high-voltage excitation parameter combination.

[0106] This embodiment discloses a method and system for detecting insulation faults in high-voltage cables. By combining multi-dimensional excitation parameter combinations with a model, it achieves accurate assessment and proactive maintenance of the insulation state of high-voltage cables. First, a multi-level high-voltage excitation parameter combination based on voltage level and voltage application frequency is used to obtain leakage current, overcoming the limitations of single-frequency or voltage excitation and enabling the stimulation of dielectric response characteristics of different insulation defects. Second, a two-dimensional dielectric spectrum is constructed using equivalent capacitance and dielectric loss factor as dual parameters, mapping high-dimensional nonlinear data into a structured spectrum, providing a suitable input format for subsequent model processing. Furthermore, the graph structure data construction based on node attributes and adjacency matrices can characterize the correlation of dielectric parameters under different excitation parameters, capturing the distribution pattern of insulation state in the voltage-frequency space. The pre-trained model integrates local and global features through a graph attention mechanism, solving the problem of traditional methods being unable to handle complex spectrum correlations. It also achieves accurate output of fault type classification and service life regression. Finally, maintenance instructions matched to the output results are used to perform maintenance on the target high-voltage cable, improving the stability of power system operation.

[0107] Example 2

[0108] Reference Figure 3 To improve the accuracy of high-voltage cable insulation fault detection by combining high-voltage frequency domain dielectric spectrum, and thus improve the stability of power system operation, this embodiment discloses a method for detecting high-voltage cable insulation faults, mainly including:

[0109] Step 301: Obtain multiple sample high-voltage cables of the same type as the target high-voltage cable, but with different types of insulation faults and different service years, and obtain multiple leakage currents of each sample high-voltage cable under different voltage levels and different voltage excitation frequencies.

[0110] In this embodiment, the main steps include collecting multiple sample high-voltage cables with different service lives and different types of insulation faults. The type of insulation fault includes the degree of aging of the insulation condition, such as new cables, lightly aged cables, moderately aged cables, and heavily aged cables. The service life can be evaluated using actual service time in years or equivalent years. For example, sample high-voltage cables with service lives at multiple time points such as new manufacturing, 0.5 years, 1 year, 2 years, 4 years, 6 years, 10 years, 15 years, 20 years, 25 years, and 30 years can be selected. It should be noted that to ensure the accuracy of insulation fault detection of the target high-voltage cable, the sample high-voltage cables and the target high-voltage cable must be of the same type. For example, if the target high-voltage cable is a typical cross-linked polyethylene insulated cable, then typical cross-linked polyethylene insulated sample cables with different service lives and different types of insulation faults should be selected.

[0111] Secondly, for each sample high-voltage cable collected, a high-voltage frequency domain dielectric spectrum test experiment was conducted on the sample high-voltage cable to collect the leakage current of the sample high-voltage cable at each voltage excitation frequency and each voltage level, so as to obtain dielectric response parameters reflecting the insulation fault and service life of the cable based on the leakage current.

[0112] In one embodiment, the range of the voltage excitation frequency is set to... ~100Hz, voltage levels including 1kV, 2kV, 5kV, 7.5kV, etc., then based on the different voltage levels corresponding to the same voltage excitation frequency and the different voltage application frequencies corresponding to the same voltage level, a range including ( Hz, 1kV), ( Hz, 2kV), ( Hz, 5kV), ( Multiple high-voltage excitation parameter combinations, including (Hz, 7kV), (100Hz, 1kV), etc.

[0113] For any given combination of high-voltage excitation parameters, an AC excitation signal corresponding to that combination of high-voltage excitation parameters is generated, and the AC excitation signal is applied to both ends of the sample high-voltage cable to obtain the leakage current of the sample high-voltage cable under each combination of high-voltage excitation parameters.

[0114] Step 302: Calculate the real part of the equivalent capacitance, the imaginary part of the equivalent capacitance, and the dielectric loss factor of the sample high-voltage cable at each voltage level and at each voltage excitation frequency based on the leakage current.

[0115] In this embodiment, the main steps are as follows: Based on the leakage current of the sample high-voltage cable at each voltage level and each voltage excitation frequency, and the voltage corresponding to the voltage level, obtain the insulation impedance of the sample high-voltage cable at each voltage level and each voltage excitation frequency; determine the angular frequency of the AC excitation signal according to the voltage excitation frequency; obtain the complex capacitance of the sample high-voltage cable at each voltage level and each voltage excitation frequency according to the insulation impedance, the angular frequency, and the voltage excitation frequency; separate the real part and the imaginary part of the equivalent capacitance from the complex capacitance; and obtain the dielectric loss factor of the sample high-voltage cable at each voltage level and each combination of voltage excitation frequencies according to the real part and the imaginary part of the equivalent capacitance.

[0116] During the measurement of the above parameters, the frequency f corresponding to the voltage excitation frequency, the voltage U corresponding to the voltage level, and the leakage current I(w) are recorded simultaneously. The insulation impedance Z and complex capacitance C* of the sample high-voltage cable are calculated, and the corresponding dielectric loss factor tanδ, and the real part C′ and imaginary part C″ of the equivalent capacitance are obtained accordingly. Specifically, the physical calculation relationship between the dielectric loss factor and the real and imaginary parts of the equivalent capacitance is as follows:

[0117]

[0118]

[0119]

[0120]

[0121]

[0122] Where ω = 2πf is the angular frequency corresponding to the voltage excitation frequency; w is the w-th sample high-voltage cable; j is a preset imaginary unit; U(w) represents the voltage corresponding to the w-th sample high-voltage cable; I(w) represents the leakage current corresponding to the w-th sample high-voltage cable; The complex capacitance of the w-th sample high-voltage cable is represented by tanδ(w); the dielectric loss factor of the w-th sample high-voltage cable is tanδ(w).

[0123] After calculating the dielectric loss factor tanδ, the real part C′ of the equivalent capacitance, and the imaginary part C″ of the equivalent capacitance, for any given sample high-voltage cable, dielectric response parameter data of the sample high-voltage cable at multiple voltage excitation frequencies and multiple voltage levels will be obtained. The data structure of the dielectric response parameter data can be represented as follows:

[0124]

[0125] in Indicates the number of voltage excitation frequencies. Indicates the number of voltage levels; the Represents the i-th voltage excitation frequency; the This represents the j-th voltage level. The dielectric loss factor of the representative high-voltage cable at the i-th voltage excitation frequency and the j-th voltage level; The real part of the equivalent capacitance of the representative high-voltage cable at the i-th voltage excitation frequency and the j-th voltage level; The dataset represents the imaginary part of the equivalent capacitance of a high-voltage cable at the i-th voltage excitation frequency and the j-th voltage level; this dataset provides a high-quality raw input foundation for subsequent spectrum construction, graph structure modeling, and state assessment.

[0126] Step 303: Using the voltage level and the voltage excitation frequency as coordinate axes, and the real part of the equivalent capacitance, the imaginary part of the equivalent capacitance, and the dielectric loss factor as coordinate values, construct a two-dimensional high-voltage dielectric spectrum corresponding to each of the sample high-voltage cables.

[0127] In this embodiment, the main steps include: after completing the measurement of the dielectric response parameter data, the collected dielectric response parameter data under multiple voltage excitation frequencies and multiple voltage levels need to be structured and organized to generate a two-dimensional high voltage dielectric spectrum, providing high-quality input for subsequent graph structure modeling and graph neural network training.

[0128] Specifically, based on the voltage excitation frequency of each sample high-voltage cable. With voltage level The equivalent capacitance real part, equivalent capacitance imaginary part, and dielectric loss factor under the parameter combination formed by each voltage excitation frequency and voltage level are constructed into the two-dimensional matrix to obtain the two-dimensional high voltage dielectric spectrum corresponding to each sample high voltage cable.

[0129] To ensure that each parameter in the two-dimensional high-voltage dielectric spectrum has uniform dimensions and distribution characteristics, the dielectric loss factor, the real part of the equivalent capacitance, and the imaginary part of the equivalent capacitance need to be normalized. Specifically, the normalization process is performed using Min-Max normalization; the calculation expression for the Min-Max normalization is as follows:

[0130]

[0131] Among them, the The values ​​represent the normalized dielectric loss factor, the real part of the equivalent capacitance, or the imaginary part of the equivalent capacitance; min(x) represents the minimum value among the dielectric loss factor, the real part of the equivalent capacitance, or the imaginary part of the equivalent capacitance; max(x) represents the maximum value among the dielectric loss factor, the real part of the equivalent capacitance, or the imaginary part of the equivalent capacitance; The dielectric loss factor, the real part of the equivalent capacitance, or the imaginary part of the equivalent capacitance before normalization are represented by the two-dimensional high-voltage dielectric spectrum after normalizing the dielectric loss factor, the real part of the equivalent capacitance, and the imaginary part of the equivalent capacitance. This spectrum can be represented as a three-channel tensor. ;

[0132] The first dimension is the number of frequency points. The second dimension is the number of voltage levels. The third dimension contains three channels, which correspond to the normalized dielectric loss factor, the real part of the equivalent capacitance, and the imaginary part, respectively.

[0133] It is important to note that during the data organization process, to ensure the consistency and comparability of the two-dimensional high-voltage dielectric spectra, the voltage excitation frequency and voltage level need to be sorted according to a standard order. The voltage excitation frequency is arranged from low to high in logarithmic step size, and the voltage level is arranged in ascending order of voltage level, in order to generate the final two-dimensional high-voltage dielectric spectrum corresponding to each sample high-voltage cable.

[0134] In this first embodiment, to improve the accuracy of insulation fault detection results for high-voltage cables, the following auxiliary characteristic indicators can be calculated based on a two-dimensional high-voltage dielectric spectrum: wherein, the multi-frequency average loss factor among the auxiliary characteristic indicators is:

[0135]

[0136] Among them, the This represents the average value of the dielectric loss factor measured at all frequency points under the j-th voltage level; This represents the dielectric loss factor measured under the combined conditions of the i-th voltage excitation frequency and the j-th voltage level.

[0137] High- and low-frequency range factor (measures the change in spectral response):

[0138]

[0139] Among them, the Represents voltage level Below, the high- and low-frequency range of the real part of the normalized capacitance is given. This range reflects the magnitude of the change of the real part of the capacitance with frequency and is used to evaluate the polarization characteristics of insulating materials at different frequencies. Represents the highest frequency point after normalization (the th The real part of the capacitance at the frequency. This represents the real part of the capacitance at the lowest frequency point (the first frequency) after normalization. Represents voltage level Below, the high- and low-frequency range of the normalized imaginary part of the capacitance is calculated. This range reflects the magnitude of the change of the imaginary part of the capacitance with frequency and is used to evaluate the polarization characteristics of the insulating material at different frequencies; It is the highest frequency point after normalization (the th The imaginary part of the capacitance at the frequency; It is the normalized imaginary part of the capacitance at the lowest frequency point (the first frequency); This represents the high-frequency range of the normalized dielectric loss factor at different voltage levels. This range reflects the magnitude of the dielectric loss factor variation with frequency and is used to directly assess the loss trend of insulating materials. Represents the highest frequency after normalization The dielectric loss factor at the point; This represents the normalized dielectric loss factor at the lowest frequency point.

[0140] These statistics can serve as additional attribute features for nodes or edges in subsequent graph neural networks, enhancing the model's ability to express insulation degradation trends. Ultimately, a two-dimensional high-voltage dielectric spectrum can be generated for each sample high-voltage cable under each test, and topological information can be embedded based on physical structures such as frequency and voltage, providing input for the next step of graph structure construction.

[0141] Step 304: Obtain the node attribute graph and adjacency matrix corresponding to the two-dimensional high voltage dielectric spectrum according to the preset node construction strategy and edge construction strategy.

[0142] In this embodiment, the main steps are as follows: for any two-dimensional high-voltage dielectric spectrum corresponding to a sample high-voltage cable, based on a preset node extraction strategy, multiple nodes are extracted from the two-dimensional high-voltage dielectric spectrum to form a node set; wherein, each node corresponds to a combination point of voltage excitation frequency and voltage level.

[0143] Specifically, in this embodiment, the joint feature points formed by the voltage excitation frequency and the voltage level are used as nodes, and a node attribute map is constructed based on this. Specifically, the two-dimensional high-voltage dielectric spectrum is mapped to the node attribute map. Where g is the node attribute graph; v is the node set, and each node represents a combination point formed by voltage excitation frequency and voltage level; Let X be the set of edges, representing the adjacency relationships between nodes; X is the node attribute matrix, recording the feature vector corresponding to each node.

[0144] Let the frequency set be The voltage set is Then the total number of nodes is N= × Each node ∈V corresponds to the voltage excitation frequency With voltage level Its attribute vector can be composed of the dielectric loss factor, the real part of the equivalent capacitance, the imaginary part of the equivalent capacitance, and the multi-frequency average loss factor and high-low frequency range factor obtained in the preferred embodiment. Specifically, the calculation expression of the attribute vector is:

[0145]

[0146] Among them, the The attribute vector represents the node corresponding to the i-th voltage excitation frequency and the j-th voltage level.

[0147] Next, according to the preset connection strategy, the physical connection relationships between each node in the node set are generated to form an edge set; wherein, the connection strategy includes a frequency connection strategy, a voltage connection strategy, and a diagonal connection strategy; wherein, the frequency connection strategy connects nodes that are at the same voltage level and are adjacent on the frequency coordinate axis; the voltage connection strategy connects nodes that are at the same voltage excitation frequency and are adjacent on the voltage coordinate axis; the diagonal connection strategy connects nodes that are adjacent on both the frequency coordinate axis and the voltage coordinate axis.

[0148] Preferably, based on a preset edge construction strategy, edges in the node attribute graph are used to describe the physical correlation and structural relationship between nodes. The following three edge connection methods are considered: frequency linking edge strategy: connecting adjacent voltage excitation frequency points under the same voltage level; voltage adjacency edge strategy: connecting nodes with adjacent voltage levels under the same voltage excitation frequency; diagonal edge strategy: connecting nodes that are adjacent on both the frequency and voltage coordinate axes to enhance the capture of the frequency-voltage joint trend.

[0149] Finally, the adjacency matrix corresponding to the two-dimensional high-voltage dielectric spectrum is generated based on the topology of the node attribute graph.

[0150] In some embodiments of this example, the adjacency matrix of the node attribute graph is used to describe the connection relationships between nodes; and the adjacency matrix is ​​constructed as follows: A frequency set and a voltage set are defined: Frequency set: F = {f1, f2, ..., f...} Voltage set: U = {U1, U2, ..., U} }; Total number of nodes N= × Next, the nodes are numbered, and the node numbering method is as follows: each node... Corresponding to and The adjacency matrix A is an N×N Boolean matrix representing whether there are edges connecting the nodes. Next, if a node... and If nodes are connected (adjacent frequencies under the same voltage), the corresponding position in the adjacency matrix is ​​assigned a value of 1; if nodes and If nodes are connected (adjacent frequencies under the same voltage), the corresponding position in the adjacency matrix is ​​assigned a value of 1; if nodes and Connect the nodes and assign a value of 1 to the corresponding positions in the adjacency matrix. This yields the adjacency matrix corresponding to the node attribute graph.

[0151] Step 305: Train the pre-constructed graph neural network model based on the node attribute graph, adjacency matrix, insulation fault type and service life of each sample high-voltage cable in the multiple sample high-voltage cables to obtain the high-voltage cable insulation fault detection model.

[0152] In this embodiment, the main steps include: to train the pre-built graph neural network model, it is necessary to label the insulation fault type and service life of each sample high-voltage cable. After completing the node attribute graph and adjacency matrix corresponding to a single sample high-voltage cable, a representative graph sample library needs to be constructed to achieve supervised learning of the model. Specifically, cable sample data with clear service life and insulation fault type are collected, and the corresponding two-dimensional high-voltage dielectric spectrum is extracted through the same high-voltage frequency domain dielectric spectrum measurement process. After normalization and feature extraction, a node attribute graph g(k) = (v(k)) is generated. The system consists of a node attribute graph g(k), X(k)), and an adjacency matrix, where k represents the k-th sample instance. Furthermore, each node attribute graph g(k) is labeled with its actual service life T(k) and insulation fault type L(k) to generate a label corresponding to each sample high-voltage cable. These labels, along with the node attribute graph and adjacency matrix, form a training set to provide a data foundation for model learning.

[0153] All training samples are organized into graph batches and their corresponding labels are loaded. Each node attribute graph contains two labels: insulation fault type and equivalent service life, which serve as the targets for classification and regression tasks, respectively.

[0154] In this first embodiment, the pre-built graph neural network model adopts a graph attention network framework, consisting of multi-layer node update units, a graph-level aggregation layer, and an output layer. The multi-layer node update units update the embedded features of each node in the node attribute graph in the following way:

[0155]

[0156] in, The feature representation of node i at layer l; The feature representation of node j at layer l; Let l be the trainable linear mapping matrix of the l-th layer; This represents a non-linear activation function pre-set in the model to improve the model's expressive power; This represents the attention weight of node j to node i in layer l, reflecting its information contribution in layer l. Specifically, in this embodiment, the multi-layer node update unit weighted and fused the feature information of the current node and its neighboring nodes, introducing the local correlation and trend of the voltage excitation frequency and voltage level of the cable. The model's expressive power is enhanced by introducing the nonlinear activation function LeakyReLU, and the importance weights of neighboring nodes are adaptively allocated through an attention mechanism, thereby improving the targeting and discriminative ability of feature extraction.

[0157] It should be noted that the attention weights This is achieved through a learnable attention scoring mechanism:

[0158]

[0159] in Let be the attention parameter vector of the l-th layer; || denotes the feature concatenation operation; LeakyReLU() is a nonlinear activation function with leakage current, used to enhance the response to negative inputs; where Let N(i) be the feature representation of node j at layer l; N(i) is the local neighborhood of node i; exp() represents the exponential function operation; The transpose of the attention parameter vector of layer l is a row vector. Through the above attention mechanism, each node can adaptively aggregate important information from neighboring nodes, overcoming the oversmoothing problem caused by fixed weight averaging in traditional GCN, and improving the ability to express local differences in the dielectric spectrum of high voltage frequency domain.

[0160] The graph-level representation vector corresponding to the node attribute graph is obtained through a graph readout function. The final representations of all nodes are aggregated into a representative feature vector of the entire graph, thus obtaining the fused global feature vector corresponding to the node attribute graph. This enables the graph neural network to perform graph-level classification and regression tasks. The process of obtaining the fused global feature vector employs an average pooling strategy.

[0161]

[0162] Where L represents the last layer of the graph neural network. Represents the set of all nodes. This represents the embedding vector of node i in the Lth (last) layer of the graph neural network. This represents the number of nodes in graph k. This represents the fused global feature vector.

[0163] To further enhance the expressive power and generalization performance of graph neural networks, an ensemble learning module is introduced to fuse the graph-level representations extracted by multiple sub-models. There are M parallel sub-models, each extracting graph representations based on different graph structure features or adjacency patterns. It employs a data-driven attention mechanism to adaptively learn the weights of each sub-model. Its aggregate expression is defined as follows:

[0164]

[0165] in The graph representation extracted for the m-th graph neural network sub-model; The weight coefficients for the m-th model reflect its importance to the final decision. The fused graph representation vector, W, As a learnable parameter, tanh() is used to increase nonlinearity to capture feature heterogeneity between different sub-models. This represents a learnable weight vector, a parameter in the attention mechanism. The superscript T indicates transpose, hence it is a row vector. This represents the atlas representation vector extracted by the j-th sub-model. Final fused representation. The original graph representation zk is replaced as the input to the output layer. The ensemble learning module integrates graph-level representations extracted from multiple sub-models. Each sub-model learns a graph representation based on different structural features or adjacency relationships, and adaptively learns the weight coefficients of each sub-model through an attention mechanism, thereby reflecting its contribution to the final decision.

[0166] The output layer of the model includes two branches: a classification branch for aging level assessment, which uses the Softmax function for three-class classification.

[0167]

[0168] The regression branch is used to assess the equivalent service life, employing a linear regression structure:

[0169] Graph-level representation vector for each sample

[0170] in , and , These are learnable parameters.

[0171] The joint loss function is defined as:

[0172]

[0173] in , For the weighting factors of the task; This represents the aging level predicted by the model for the k-th sample (i.e., the spectrum of the k-th cable); This represents the true aging level of the k-th sample. The equivalent service life predicted by the model for the k-th sample; The true equivalent service life of each sample; C represents the total number of categories in the classification task; This represents the true aging level of the k-th sample. During the training of the model, end-to-end training is performed using the Adam optimizer. Modeling is complete once the loss function converges, resulting in the trained high-voltage cable insulation fault detection model.

[0174] Step 306: Input the node attribute graph and adjacency matrix corresponding to the target high-voltage cable into the high-voltage cable insulation fault detection model to obtain the insulation fault type and service life of the target high-voltage cable, and control the preset maintenance equipment to perform maintenance on the target high-voltage cable according to the insulation fault type and service life.

[0175] In this embodiment, the step mainly includes: testing the target high-voltage cable according to the same high-voltage frequency domain dielectric spectrum testing experimental procedure as the sample high-voltage cable to obtain the leakage current corresponding to the target high-voltage cable. Specifically, based on multiple voltage application frequencies and multiple voltage levels, multiple high-voltage excitation parameter combinations are generated, and the leakage current of the target high-voltage cable under each of the high-voltage excitation parameter combinations is obtained. Then, the real part of the equivalent capacitance, the imaginary part of the equivalent capacitance, and the dielectric loss factor corresponding to any one of the leakage currents are obtained.

[0176] Using the voltage level and the voltage excitation frequency as coordinate axes, and the real part of the equivalent capacitance, the imaginary part of the equivalent capacitance, and the dielectric loss factor as coordinate values, a two-dimensional high-voltage dielectric spectrum corresponding to the target high-voltage cable is constructed. The node attribute graph and adjacency matrix corresponding to the two-dimensional high-voltage dielectric spectrum are obtained according to the same node construction strategy and edge construction strategy as those in the sample high-voltage cable.

[0177] The node attribute graph and adjacency matrix are input into the high-voltage cable insulation fault detection model obtained after training, so as to output the detection results of the insulation fault type and service life of the target high-voltage cable. Then, based on the detection results, the fault operation and maintenance operation instructions corresponding to the target high-voltage cable are generated, and the preset operation and maintenance equipment is controlled to perform maintenance operations on the target high-voltage cable based on the fault operation and maintenance operation instructions.

[0178] Preferably, for a target high-voltage cable that has been operating continuously in harsh environments such as high temperature and high humidity for 3 years, its dielectric response spectrum is measured and input into the model. The model outputs that its insulation fault type corresponds to a severe aging fault, with a corresponding equivalent service life assessment value of 10.2 years. This result indicates that the degree of cable insulation degradation has significantly exceeded the expected value of normal service life, suggesting that the current cable has a relatively serious potential failure risk. At this time, the system automatically sends the maintenance work order corresponding to the detection result to the preset maintenance equipment and simultaneously controls the parallel backup cable to be put into operation to avoid fault accidents caused by insulation failure.

[0179] This embodiment provides a method for detecting insulation faults in high-voltage cables. First, cable samples with different service years are selected to conduct systematic high-voltage frequency domain dielectric spectrum experiments. By applying multi-level high-voltage AC excitation to each sample and collecting response parameters such as dielectric loss factor, real and imaginary parts of capacitance at multiple frequency points, a two-dimensional high-voltage frequency domain dielectric spectrum is constructed, and a standard sample database is established based on this spectrum. Then, a graph neural network framework based on graph attention mechanism is used to train the database, establishing an evaluation model for assessing insulation fault types and equivalent service years. In the application stage, the high-voltage frequency domain dielectric spectrum of the target high-voltage cable is input into the model to automatically classify and judge its insulation faults and output the equivalent service year result. This method has advantages such as non-contact operation, high sensitivity, and clear evaluation criteria, making it suitable for online detection and intelligent operation and maintenance management scenarios for cable insulation conditions.

[0180] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for detecting insulation faults in high-voltage cables, characterized in that, include: The target high-voltage cable is subjected to multi-stage high-voltage excitation to obtain the leakage current of the target high-voltage cable under different combinations of high-voltage excitation parameters; wherein, the high-voltage excitation parameter combination includes voltage level and voltage excitation frequency; Based on the leakage current, obtain the equivalent capacitance and dielectric loss factor of the target high-voltage cable under each of the high-voltage excitation parameter combinations; Using the voltage level and the voltage excitation frequency as coordinate axes, and the equivalent capacitance and the dielectric loss factor as coordinate values, a two-dimensional high-voltage dielectric spectrum corresponding to the target high-voltage cable is constructed. The node attribute graph and adjacency matrix corresponding to the two-dimensional high-voltage dielectric spectrum are obtained according to a preset node construction strategy and edge construction strategy. Specifically, based on a preset node extraction strategy, multiple nodes are extracted from the two-dimensional high-voltage dielectric spectrum to form a node set. Each node corresponds to a combination of voltage excitation frequency and voltage level. According to a preset connection strategy, physical connection relationships between the nodes in the node set are generated to form an edge set. The connection strategies include frequency connection strategy, voltage connection strategy, and diagonal connection strategy. The frequency connection strategy connects nodes at the same voltage level along the frequency coordinate axis. The voltage connection strategy connects nodes that are adjacent on the voltage coordinate axis at the same voltage excitation frequency; the diagonal connection strategy connects nodes that are adjacent on both the frequency coordinate axis and the voltage coordinate axis; the normalized real part, normalized imaginary part, and normalized dielectric loss factor of each node are obtained from the two-dimensional high-voltage dielectric spectrum as the feature vector of the node; a node attribute graph corresponding to the two-dimensional high-voltage dielectric spectrum is generated based on the feature vector, the edge set, and the node set; an adjacency matrix corresponding to the two-dimensional high-voltage dielectric spectrum is generated based on the topology of the node attribute graph. The adjacency matrix and the node attribute graph are input into a pre-trained high-voltage cable insulation fault detection model to output the detection results of the insulation fault type and service life of the target high-voltage cable. Based on the detection results, a fault maintenance operation instruction corresponding to the target high-voltage cable is generated, and based on the fault maintenance operation instruction, a preset maintenance equipment is controlled to perform maintenance operations on the target high-voltage cable.

2. The method for detecting insulation faults in high-voltage cables according to claim 1, characterized in that, The process of applying multi-stage high-voltage excitation to the target high-voltage cable to obtain the leakage current of the target high-voltage cable under different combinations of high-voltage excitation parameters includes: Multiple sets of high-voltage excitation parameter combinations are generated based on multiple preset voltage excitation frequencies and multiple preset voltage levels; wherein, different high-voltage excitation parameter combinations have different voltage excitation frequencies or different voltage levels. For any set of the high-voltage excitation parameter combinations: An AC excitation signal corresponding to the high-voltage excitation parameter combination is generated, and the AC excitation signal is applied to both ends of the target high-voltage cable to measure the leakage current of the target high-voltage cable under the high-voltage excitation parameter combination.

3. The method for detecting insulation faults in high-voltage cables according to claim 2, characterized in that, The step of obtaining the equivalent capacitance and dielectric loss factor of the target high-voltage cable under each combination of high-voltage excitation parameters based on the leakage current includes: For any set of the high-voltage excitation parameter combinations: Based on the voltage corresponding to the voltage level in the high-voltage excitation parameter combination and the leakage current of the target high-voltage cable under the high-voltage excitation parameter combination, the insulation impedance of the target high-voltage cable under the high-voltage excitation parameter combination is obtained. The angular frequency of the AC excitation signal is determined based on the voltage excitation frequency in the high-voltage excitation parameter combination. Based on the insulation impedance, the angular frequency, and the voltage excitation frequency, the complex capacitance of the target high-voltage cable under the given combination of high-voltage excitation parameters is obtained. Separate the real part and imaginary part of the equivalent capacitance of the target high-voltage cable under the given high-voltage excitation parameter combination from the complex capacitance; Based on the real part and the imaginary part of the equivalent capacitance, the dielectric loss factor of the target high-voltage cable under the given high-voltage excitation parameter combination is obtained.

4. The method for detecting insulation faults in high-voltage cables according to claim 3, characterized in that, The step of obtaining the equivalent capacitance and dielectric loss factor of the target high-voltage cable under each combination of high-voltage excitation parameters based on the leakage current includes: Obtain a first ratio of the voltage divided by the leakage current, use the first ratio as the insulation resistance, and obtain a first reciprocal of the insulation resistance; Obtain the frequency product of the preset imaginary unit and the angular frequency, and obtain the second reciprocal of the frequency product, so that the product of the first reciprocal and the second reciprocal can be used as the complex capacitance; Obtain a second ratio of the imaginary part of the equivalent capacitance to the real part of the equivalent capacitance, and use the second ratio as the dielectric loss factor of the target high-voltage cable under the high-voltage excitation parameter combination.

5. The method for detecting insulation faults in high-voltage cables according to claim 3, characterized in that, The construction of a two-dimensional high-voltage dielectric spectrum corresponding to the target high-voltage cable, using the voltage level and voltage excitation frequency as coordinate axes and the equivalent capacitance and dielectric loss factor as coordinate values, includes: For any set of the high-voltage excitation parameter combinations: The real part of the equivalent capacitance, the imaginary part of the equivalent capacitance, and the dielectric loss factor corresponding to the high-voltage excitation parameter combination are normalized to obtain the normalized real part of the equivalent capacitance, the normalized imaginary part of the equivalent capacitance, and the normalized dielectric loss factor corresponding to the high-voltage excitation parameter combination. According to the preset logarithmic step size and sorting rules from low to high, the multiple voltage excitation frequencies in the multiple high voltage excitation parameter combinations are arranged to obtain the frequency coordinate axis; According to a preset ascending order of voltage levels, the voltage levels in the multiple combinations of high-voltage excitation parameters are arranged to obtain a voltage coordinate axis. A two-dimensional matrix is ​​constructed based on the frequency coordinate axis and the voltage coordinate axis. The real part of the normalized equivalent capacitance, the imaginary part of the normalized equivalent capacitance, and the normalized dielectric loss factor corresponding to each combination of high-voltage excitation parameters are embedded into the two-dimensional matrix to obtain the two-dimensional high-voltage dielectric spectrum corresponding to the target high-voltage cable.

6. The method for detecting insulation faults in high-voltage cables according to claim 1, characterized in that, The step of inputting the adjacency matrix and the node attribute graph into a pre-trained high-voltage cable insulation fault detection model to output the detection results of the insulation fault type and service life of the target high-voltage cable includes: The adjacency matrix and the node attribute graph are synchronously input into the pre-trained high-voltage cable insulation fault detection model so as to extract the embedding features corresponding to each node in the node attribute graph through the high-voltage cable insulation fault detection model. For any node structure relationship preset in the high-voltage cable insulation fault detection model: The multiple embedded features obtained are averaged according to the node structure relationship to obtain the global feature vector corresponding to the node structure relationship; The multiple global feature vectors are adaptively weighted and fused by an attention mechanism pre-set in the high-voltage cable insulation fault detection model to obtain a fused global feature vector. The fused global feature vector is regressed and classified according to the regression classification function preset in the high-voltage cable insulation fault detection model, and the insulation status assessment result and equivalent service life of the target high-voltage cable are obtained and output.

7. The method for detecting insulation faults in high-voltage cables according to claim 6, characterized in that, The step of synchronously inputting the adjacency matrix and the node attribute graph into a pre-trained high-voltage cable insulation fault detection model, so as to extract the embedding features corresponding to each node in the node attribute graph through the high-voltage cable insulation fault detection model, includes: For any node in the node attribute graph: Based on the adjacency matrix, the high-voltage cable insulation fault detection model identifies multiple adjacent nodes corresponding to the node, and extracts the first feature information of the node and the second feature information of each of the adjacent nodes. The weights corresponding to each of the second feature information are assigned by the attention mechanism preset in the high-voltage cable insulation fault detection model. Based on the weights, the first feature information and multiple second feature information are weighted and fused using the high-voltage cable insulation fault detection model to obtain the embedded feature corresponding to the node.

8. A detection system for insulation faults in high-voltage cables, characterized in that, It includes a multi-level high-voltage excitation module, a fault medium extraction module, a spectrum mapping module, a node attribute extraction module, an insulation fault detection module, and a cable fault operation and maintenance module; The multi-stage high-voltage excitation module is used to perform multi-stage high-voltage excitation on the target high-voltage cable to obtain the leakage current of the target high-voltage cable under different combinations of high-voltage excitation parameters; wherein, the high-voltage excitation parameter combination includes voltage level and voltage excitation frequency; The fault medium extraction module is used to obtain the equivalent capacitance and dielectric loss factor of the target high-voltage cable under each combination of high-voltage excitation parameters based on the leakage current. The spectrum mapping module is used to construct a two-dimensional high-voltage dielectric spectrum corresponding to the target high-voltage cable, with the voltage level and the voltage excitation frequency as coordinate axes and the equivalent capacitance and the dielectric loss factor as coordinate values. The node attribute extraction module is used to obtain the node attribute graph and adjacency matrix corresponding to the two-dimensional high-voltage dielectric spectrum according to a preset node construction strategy and edge construction strategy. Specifically, based on the preset node extraction strategy, multiple nodes are extracted from the two-dimensional high-voltage dielectric spectrum to form a node set; each node corresponds to a combination of voltage excitation frequency and voltage level. According to a preset connection strategy, physical connection relationships between nodes in the node set are generated to form an edge set; the connection strategy includes a frequency connection strategy, a voltage connection strategy, and a diagonal connection strategy; the frequency connection strategy connects nodes at the same voltage level. The voltage connection strategy connects nodes that are adjacent on the frequency axis; the diagonal connection strategy connects nodes that are adjacent on both the frequency axis and the voltage axis; the normalized real part, normalized imaginary part, and normalized dielectric loss factor of each node are obtained from the two-dimensional high-voltage dielectric spectrum as feature vectors of the node; a node attribute graph corresponding to the two-dimensional high-voltage dielectric spectrum is generated based on the feature vectors, the edge set, and the node set; an adjacency matrix corresponding to the two-dimensional high-voltage dielectric spectrum is generated based on the topology of the node attribute graph. The insulation fault detection module is used to input the adjacency matrix and the node attribute graph into the pre-trained high-voltage cable insulation fault detection model, so as to output the detection results of the insulation fault type and service life of the target high-voltage cable. The cable fault maintenance module is used to generate fault maintenance operation instructions corresponding to the target high-voltage cable based on the detection results, and control the preset maintenance equipment to perform maintenance operations on the target high-voltage cable based on the fault maintenance operation instructions.

9. A high-voltage cable insulation fault detection system according to claim 8, characterized in that, The multi-stage high-voltage excitation module includes a parameter combination unit and an excitation unit; The parameter combination unit is used to generate multiple sets of high-voltage excitation parameter combinations based on multiple preset voltage excitation frequencies and multiple preset voltage levels; wherein, different high-voltage excitation parameter combinations have different voltage excitation frequencies or different voltage levels; The excitation unit is used to generate an AC excitation signal corresponding to any set of high-voltage excitation parameter combinations, and apply the AC excitation signal to both ends of the target high-voltage cable to measure the leakage current of the target high-voltage cable under the high-voltage excitation parameter combination.

Images