A method for detecting defects in the lead sealing layer of a high-voltage cable

Abstract

The present invention relates to the field of high-voltage cable defect detection, and in particular to a method for detecting defects in the lead sealing layer of a high-voltage cable. It includes: collecting real-time parameters of the cable through an embedded sensor network; scanning the lead sealing layer using phased array ultrasonic technology to extract key frequency domain features; reconstructing the three-dimensional morphology of defects based on an inverse scattering matrix analysis algorithm; generating a defect model in combination with a deep learning algorithm to accurately locate defects; generating an optimal repair strategy using a Markov random field model; and performing real-time feedback and remote monitoring through the Internet of Things technology to achieve dynamic adjustment and intelligent management. The present invention can effectively improve the accuracy and comprehensiveness of defect detection in the lead sealing layer of high-voltage cables, and provide an intelligent repair solution that adapts to the complex environment of long-term operation of high-voltage cables, significantly enhancing the maintenance and management efficiency of high-voltage cables.

Description

Technical Field

[0001] The invention relates to the field of high-voltage cable defect detection, and in particular to a method for detecting defects in a lead sealing layer of a high-voltage cable. Background Art

[0002] In the manufacturing process of high-voltage cables, the lead sealing layer plays an important sealing role, which can prevent moisture and other contaminants from entering the cable, thereby ensuring the long-term stable operation of the cable. However, during the production and installation process, defects may occur in the lead sealing layer, such as cracks, pores or uneven thickness, which may cause the cable performance to deteriorate or even cause failures. Therefore, how to effectively detect the defects of the lead sealing layer has become a key issue in the quality control of high-voltage cables. At present, there are still the following problems: the existing ultrasonic detection technology usually adopts a fixed frequency and angle scanning method when detecting defects in the cable lead sealing layer, which is easily interfered by signal noise, resulting in inaccurate defect positioning and insufficient detection accuracy; the existing technology is difficult to comprehensively consider the influence of multiple parameters, lacks fusion analysis of multimodal data, resulting in insufficient comprehensive description of defects, and it is difficult to accurately generate a three-dimensional model of defects; the existing technology relies on manual experience for repair decisions, lacks intelligent optimization means, and the formulation of repair strategies is inefficient, and it is easy to make misjudgments, resulting in increased repair costs and reduced efficiency. Summary of the invention

[0003] To solve the above problems, the present invention provides a method for detecting defects in the lead layer of a high-voltage cable, which solves the problem of how to improve the accuracy of detecting defects in the lead layer of a high-voltage cable, and optimizes the repair strategy through intelligent means to reduce costs and improve efficiency, thereby enhancing the maintenance and management efficiency of the high-voltage cable.

[0004] To achieve the above object, the technical solution adopted by the present invention is:

[0005] A method for detecting defects in a high-voltage cable lead sealing layer comprises the following steps:

[0006] S1: Based on the real-time operating status of the high-voltage cable, real-time parameters of the high-voltage cable are collected through an embedded sensor network, wherein the real-time parameters include temperature, pressure, current and voltage of the high-voltage cable;

[0007] S2: Use phased array ultrasonic technology to scan the lead sealing layer of the high-voltage cable, dynamically adjust the frequency and angle, and extract the key frequency domain features of the ultrasonic signal;

[0008] S3: Based on key frequency domain characteristics and real-time parameters, the inverse scattering matrix analysis algorithm is used to analyze the scattering and reflection characteristics of electromagnetic waves inside the lead sealing layer, reconstruct the three-dimensional morphology of the defect, and form a complete defect feature through multi-modal data fusion;

[0009] S4: Combine multimodal fusion data and use deep learning algorithms to analyze data and generate defect models, accurately locate the position, depth and size of defects, and form a complete defect distribution map;

[0010] S5: inputting the defect distribution graph into a Markov random field model, and generating an optimal repair strategy through probabilistic reasoning and optimization of the graph structure;

[0011] S6: Dynamically adjust the high-voltage cables based on the optimal repair strategy, and perform real-time feedback and remote monitoring of defect detection through IoT technology.

[0012] Furthermore, the phased array ultrasonic technology includes using a multi-frequency ultrasonic transmitting array and a receiving array to control ultrasonic waves of different frequencies to scan the lead sealing layer at multiple angles and depths.

[0013] Furthermore, the S3 specifically includes the following steps:

[0014] S31: Based on the key frequency domain features, an inverse scattering matrix is ​​constructed and the matrix is ​​subjected to bandpass filtering;

[0015] S32: Using the inverse scattering matrix analysis algorithm, the scattering coefficient of each frequency band signal is calculated, the multiple scattering paths in the cable lead sealing layer are identified, and the preliminary location of the defect is determined;

[0016] S33: Time-reversing the multiple scattering paths, reconstructing the scattering scene inside the lead sealing layer through the back propagation algorithm, and accurately locating the three-dimensional coordinates of the defects;

[0017] S34: Match and verify the back propagation results of the multiple scattering paths with the key frequency domain features, and generate a three-dimensional defect feature map containing position and morphology information.

[0018] Furthermore, the formula of the inverse scattering matrix analysis algorithm is as follows:

[0019] ,

[0020] in, It represents the defect distribution function from the i-th ultrasonic transmitter to the j-th receiver at frequency f, that is, the defect characteristics inside the lead sealing layer; It represents the amplitude factor of the kth scattering path, reflecting the attenuation and energy loss on the path in the lead sealing layer of the high-voltage cable; represents the time delay of the kth scattering path; represents the propagation distance of the kth scattering path, which can be obtained by scanning the lead sealing layer at multiple angles and depths using multi-frequency ultrasonic waves; represents the scattering attenuation index; represents the phase matrix element in the transfer matrix at frequency f; represents the noise matrix element at frequency f.

[0021] Furthermore, the S4 specifically includes the following steps:

[0022] S41: Based on the extracted multimodal fusion data, an autoencoder is used to reduce the dimension and extract features of the defect data, and the potential features of the defects are preliminarily identified;

[0023] S42: Use graph neural networks to model the graph structure of defect features, capture the complex spatial relationship of defects, and combine generative adversarial networks to perform adversarial optimization on the three-dimensional morphology of defects, and refine the defect edges and depths through adversarial training;

[0024] S43: Generate a final defect distribution map based on the optimized three-dimensional morphology reconstruction.

[0025] Furthermore, the S42 specifically includes the following steps:

[0026] A dynamic graph structure model based on multimodal data is constructed. The defect characteristics of the lead sealing layer of the high-voltage cable are used as nodes in the graph structure. The edges between the nodes represent the potential causal relationship, and the edge weights are adjusted to adapt to changes in the operating environment.

[0027] Adopting adaptive graph neural network, it captures the complex spatial characteristics of defects through self-supervised learning, identifies new defects through topological learning, and updates the graph structure in real time;

[0028] Generative adversarial networks are introduced to use multi-channel parallel computing to model the three-dimensional morphology of defects layer by layer;

[0029] Based on the optimized 3D morphology and combined with real-time high-voltage cable operating parameters, a time-dynamic defect distribution map is generated.

[0030] Furthermore, the S5 specifically includes the following steps:

[0031] S51: discretize the defect distribution map data and construct a state space model including different areas of the high-voltage cable lead sealing layer;

[0032] S52: Based on the state space model, the Markov random field model is used to map the spatial position, depth and correlation of the defect with the adjacent area into random variables, and a state transition matrix is ​​constructed to define the transition probability between states;

[0033] S53: Based on the state transfer matrix, applying optimal control theory, evaluating different repair strategies, calculating the expected repair effect under each state through a dynamic programming algorithm, and identifying the optimal repair path;

[0034] S54: Combined with the real-time operating parameters of the cable, a multi-scale hierarchical optimization algorithm is used to perform multiple rounds of iterative optimization on the optimal repair strategy, and finally a high-voltage cable lead layer defect repair plan based on the optimal control sequence is output, and the repair plan is dynamically adjusted through real-time monitoring and feedback mechanism.

[0035] Furthermore, the formula of the Markov random field model is as follows:

[0036] ,

[0037] in, represents the minimum expected repair cost under state s, where state s represents the defect state of different areas of the lead sealing layer of the high-voltage cable; a represents the repair action that can be taken under state s; represents the set of feasible repair actions in state s; represents the immediate repair cost of performing action a in state s, including material consumption and repair time; represents the discount factor, which is used to weigh the relationship between the immediate repair cost and the future repair cost, and is set according to the operating life of the cable and the maintenance plan; Indicates that after executing action a in state s, it transfers to state The probability of; S represents the set of all states, including all states that appear after the sealing layer defect is repaired.

[0038] Furthermore, the Internet of Things technology specifically combines edge computing and 5G communication technology to perform real-time data processing and analysis at edge nodes close to high-voltage cables, and achieve instant feedback and remote control of defect detection results.

[0039] The beneficial effects of the present invention are:

[0040] The present invention collects real-time parameters of high-voltage cables through embedded sensor networks, which can reflect the operating status of cables in real time, and provide accurate basic data for subsequent ultrasonic detection and defect analysis, thereby improving the accuracy of detection. By using phased array ultrasonic technology, the lead sealing layer is dynamically scanned at multiple angles and multiple frequencies, and the scattering and reflection characteristics inside the lead sealing layer are analyzed in combination with the inverse scattering matrix analysis algorithm, so that the three-dimensional morphology of the defect can be reconstructed with high precision, ensuring that the shape, size and position of the defect are fully and accurately characterized. By combining multimodal data fusion with deep learning algorithms, the characteristic information of the defect can be fully extracted and analyzed to form an accurate defect model, which helps to accurately locate the position, depth and size of the defect, and ensure the high precision and reliability of the detection results. The Markov random field model is introduced, and the optimal repair strategy can be generated based on the defect distribution through the probabilistic reasoning and optimization of the graph structure, thereby providing intelligent decision support for the maintenance of high-voltage cables, reducing repair costs and improving maintenance efficiency. Through the Internet of Things technology, the detection system can feedback the detection results in real time and support remote monitoring, ensuring that the defects of the high-voltage cable can be discovered and processed in the first time, greatly improving the safety and reliability of cable operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 The present invention is a schematic flow chart of a method for detecting defects in a high-voltage cable lead sealing layer.

[0042] Figure 2 It is a flowchart of specific steps of step S3 provided in one embodiment of the present invention.

[0043] Figure 3 It is a flowchart of specific steps of step S5 provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0044] See also Figures 1 to 3 As shown, the present invention relates to a method for detecting defects in a lead sealing layer of a high-voltage cable.

[0045] Example

[0046] A method for detecting defects in a high-voltage cable lead sealing layer comprises the following steps:

[0047] S1: Based on the real-time operating status of the high-voltage cable, real-time parameters of the high-voltage cable are collected through an embedded sensor network, wherein the real-time parameters include temperature, pressure, current and voltage of the high-voltage cable;

[0048] S2: Use phased array ultrasonic technology to scan the lead sealing layer of the high-voltage cable, dynamically adjust the frequency and angle, and extract the key frequency domain characteristics of the ultrasonic signal; the phased array ultrasonic technology includes using a multi-frequency ultrasonic transmitting array and a receiving array to control ultrasonic waves of different frequencies to scan the lead sealing layer at multiple angles and depths.

[0049] In some embodiments, a phased array ultrasonic transmitting array and receiving array are arranged on the outer surface of the cable along the cable axis. The transmitting array includes a plurality of independent ultrasonic transmitting units, each of which can transmit ultrasonic pulses at different frequencies. The receiving array includes a plurality of independent receiving units, which can capture the reflected ultrasonic signals with high sensitivity. During the scanning process, the transmitting frequency and transmitting angle of the ultrasonic wave are dynamically adjusted according to the physical properties of the cable (such as the thickness and material properties of the lead sealing layer). The specific operation includes: by changing the transmitting frequency of the ultrasonic wave, the wavelength can be controlled, thereby affecting the penetration depth of the ultrasonic wave in the lead sealing layer. Low-frequency ultrasonic waves have stronger penetration ability and are suitable for detecting deeper defects, while high-frequency ultrasonic waves are suitable for detecting subtle defects on the surface or near the surface. Using phased array technology, the transmitting array can transmit ultrasonic pulses at different angles. By adjusting the transmitting angle, the system can scan the lead sealing layer from multiple directions, thereby capturing defect information more comprehensively. The angle adjustment range is usually between 0° and 90°, which is achieved by changing the phase difference of the transmitting unit.

[0050] In actual operation, the system will first perform a global scan to obtain the preliminary outline and structural information of the lead sealing layer. Then, it will perform a focused scan on the areas where defects may exist. When performing a focused scan, the system will select the appropriate frequency and angle and perform multiple repeated scans to ensure that enough reflection signals are obtained.

[0051] Multi-angle scanning: In order to obtain the complete defect morphology, the system will scan the same position from different angles. Ultrasonic waves at different angles can penetrate different parts of the lead sealing layer and provide different perspective information about the defect. This information can be used for subsequent 3D reconstruction.

[0052] Multi-depth scanning: By adjusting the transmission frequency and the sensitivity of the receiving array, the system can scan the different depths of the lead sealing layer layer by layer, from the surface to the inner layer, ensuring that any potential defects can be detected.

[0053] After each ultrasonic pulse is transmitted, the receiving array captures the signal reflected from the lead sealing layer. The received signal usually contains multiple echoes, each echo corresponding to a reflecting surface at a different depth or position. The system performs time domain and frequency domain analysis on these echo signals to extract key frequency domain features, such as echo amplitude, delay time, and spectral characteristics.

[0054] Time domain analysis: By analyzing the time delay of the signal, the distance and depth of the reflection point can be determined, thereby preliminarily judging the location of the defect.

[0055] Frequency domain analysis: Frequency domain analysis focuses on the spectral characteristics of the signal, such as the frequency content and amplitude changes of the signal. Through frequency domain analysis, changes in ultrasonic attenuation, scattering and reflection characteristics caused by defects can be identified.

[0056] S3: Based on key frequency domain characteristics and real-time parameters, the inverse scattering matrix analysis algorithm is used to analyze the scattering and reflection characteristics of electromagnetic waves inside the lead sealing layer, reconstruct the three-dimensional morphology of the defect, and form a complete defect feature through multi-modal data fusion;

[0057] Wherein, the S3 specifically includes the following steps:

[0058] S31: Based on the key frequency domain features, an inverse scattering matrix is ​​constructed and the matrix is ​​subjected to bandpass filtering;

[0059] S32: Using the inverse scattering matrix analysis algorithm, the scattering coefficient of each frequency band signal is calculated, the multiple scattering paths in the cable lead sealing layer are identified, and the preliminary location of the defect is determined;

[0060] Specifically, the signal of each frequency band in the inverse scattering matrix is ​​analyzed. The inverse scattering matrix contains information about the interaction between electromagnetic waves and the internal structure of the lead sealing layer at different frequencies. For each frequency band, its scattering coefficient is calculated, that is, the ratio of reflection and transmission generated by the electromagnetic wave when passing through the material interface. Specifically, the inverse scattering matrix is ​​processed using matrix decomposition techniques (such as singular value decomposition SVD or least squares method LSM) to separate the effective signal from the background noise to ensure the accuracy of the calculation of the scattering coefficient. The scattering coefficient is compared and analyzed with the material properties of the lead sealing layer to detect any abnormal scattering behavior. Abnormal scattering usually manifests as higher reflection or lower transmission than expected, which may indicate the existence of inhomogeneity or defects inside the material. By statistically analyzing the scattering coefficient of each frequency band, the intensity distribution diagram of the electromagnetic wave when it propagates inside the lead sealing layer can be drawn, which provides an intuitive basis for identifying multiple scattering paths.

[0061] Multiple scattering path identification first performs pulse compression processing on the scattered signals of each frequency band in the time domain to improve the time resolution of the signal. This helps to separate multiple scattering events. Then, a delay estimation algorithm (such as a cross-correlation algorithm or a matched filter algorithm) is used to calculate the propagation time of each scattering path, thereby identifying the relative delay of different paths. A multiple scattering path map is constructed by comprehensively analyzing the scattering paths of different frequency bands. The map shows all possible paths that the electromagnetic wave passes through in the lead sealing layer, and marks the relative intensity and relative position of each path.

[0062] Combined with the multiple scattering path map and the scattering coefficient distribution map, the preliminary position of the defect is determined. Use clustering algorithms (such as K-means clustering or density peak clustering) to group similar scattering paths and cluster paths belonging to the same defect together. The location of each cluster center is the preliminary location of the defect. These locations will be further optimized and accurately located in subsequent steps.

[0063] S33: Time-reversing the multiple scattering paths, reconstructing the scattering scene inside the lead sealing layer through the back propagation algorithm, and accurately locating the three-dimensional coordinates of the defects;

[0064] It should be noted that time reversal technology is a method of reconstructing the scattering scene by propagating the received scattering signal back to the source. First, each scattering path is time-reversed using the previously calculated scattering coefficient. Specifically, the received scattering signal is transformed through time reversal and propagated back to the scattering source of the lead sealing layer. Through this process, the scattering scene inside the lead sealing layer can be reconstructed. Then, the back-propagation algorithm is used to gradually converge the propagation path of the scattering signal, and finally accurately locate the three-dimensional coordinates of the defect. This process usually combines the scattering paths of multiple frequency bands for comprehensive analysis to improve positioning accuracy.

[0065] S34: Match and verify the back propagation results of the multiple scattering paths with the key frequency domain features, and generate a three-dimensional defect feature map containing position and morphology information.

[0066] Furthermore, the formula of the inverse scattering matrix analysis algorithm is as follows:

[0067] ,

[0068] in, It represents the defect distribution function from the i-th ultrasonic transmitter to the j-th receiver at frequency f, that is, the defect characteristics inside the lead sealing layer; It represents the amplitude factor of the kth scattering path, reflecting the attenuation and energy loss on the path in the lead sealing layer of the high-voltage cable; represents the time delay of the kth scattering path, the signal propagation time delay affected by the real-time parameters of the high-voltage cable (such as temperature, pressure, current and voltage); represents the propagation distance of the kth scattering path, which can be obtained by scanning the lead sealing layer at multiple angles and depths using multi-frequency ultrasonic waves; Represents the scattering attenuation index, which is usually determined based on the material properties of the lead sealing layer of the high-voltage cable; It means that at frequency f, the phase matrix elements in the transfer matrix reflect the phase information of the signal from the ultrasonic transmitter to the receiver, which is affected by the internal structure of the high-voltage cable lead sealing layer and the real-time operating status; It represents the noise matrix element at frequency f, reflecting the noise interference that may exist in the detection process of the high-voltage cable lead sealing layer, including the measurement error caused by the real-time parameter changes of the cable.

[0069] S4: Combine multimodal fusion data and use deep learning algorithms to analyze data and generate defect models, accurately locate the position, depth and size of defects, and form a complete defect distribution map;

[0070] Wherein, the S4 specifically includes the following steps:

[0071] S41: Based on the extracted multimodal fusion data, an autoencoder is used to reduce the dimension and extract features of the defect data, and the potential features of the defects are preliminarily identified;

[0072] Specifically, a deep autoencoder model is constructed. The autoencoder consists of an encoder and a decoder, which is used to reduce the dimensionality of high-dimensional multimodal data. The encoder part consists of multiple convolutional layers and pooling layers, which extract low-dimensional feature representations of the data layer by layer; the decoder part reconstructs the reduced-dimensional data into its original form through a deconvolution layer to ensure that the extracted features retain as much original information as possible. During the training process, multimodal data is input, and the model automatically adjusts the weight parameters to minimize the reconstruction error. Through training, the autoencoder can learn the potential structure and key features of the data. The low-dimensional feature representation finally output retains the main information of the data, eliminates noise and redundancy, and forms a preliminary defect feature representation.

[0073] S42: Use graph neural networks to model the graph structure of defect features, capture the complex spatial relationship of defects, and combine generative adversarial networks to perform adversarial optimization on the three-dimensional morphology of defects, and refine the defect edges and depths through adversarial training;

[0074] S43: Generate a final defect distribution map based on the optimized three-dimensional morphology reconstruction.

[0075] Furthermore, the S42 specifically includes the following steps:

[0076] A dynamic graph structure model based on multimodal data is constructed. The defect characteristics of the lead sealing layer of the high-voltage cable are used as nodes in the graph structure. The edges between the nodes represent the potential causal relationship, and the edge weights are adjusted to adapt to changes in the operating environment.

[0077] Adopting adaptive graph neural network, it captures the complex spatial characteristics of defects through self-supervised learning, identifies new defects through topological learning, and updates the graph structure in real time;

[0078] It should be noted that an adaptive graph neural network (GNN) is constructed to model the defect features more deeply in the graph structure. GNN can capture the complex spatial relationship between defect features through the information propagation mechanism between nodes. The key to the design of adaptive GNN is that it can automatically adjust its internal structure and parameters according to the characteristics of the input data to adapt to different operating environments and defect feature distributions. The training of adaptive GNN adopts a self-supervised learning method and does not rely on manually annotated data. By designing an appropriate loss function, GNN can autonomously learn the relationship between nodes and edges during the training process. For example, GNN can learn how to effectively propagate information in the graph structure by minimizing the node feature prediction error. In addition, by reconstructing defect features and the relationship between them, GNN can also identify and express complex spatial patterns. In order to deal with new defects that may appear in high-voltage cables, GNN combines topological learning methods to update the graph structure in real time. Topological learning identifies newly emerged nodes (i.e., new defects) and their association with existing nodes by analyzing the global morphology and local connection patterns of the graph structure. After identifying new defects, GNN will automatically add them to the existing graph structure and adjust the corresponding edges and weights to ensure the integrity and accuracy of the graph structure.

[0079] Generative adversarial networks are introduced to use multi-channel parallel computing to model the three-dimensional morphology of defects layer by layer;

[0080] It should be noted that the Generative Adversarial Network (GAN) consists of a generator and a discriminator. The generator is responsible for generating the three-dimensional morphology of the defect, while the discriminator evaluates the authenticity of the generated result. In this process, adversarial training between the generator and the discriminator helps to improve the generation ability of the model. Specifically, using a multi-channel parallel computing architecture, the generator generates different features of the defect, such as edges, depth, and surface details, through different channels. Each channel uses different convolution kernels and hierarchical structures to specialize in different types of morphological features. For example, one channel may focus on generating the edges of the defect, while another channel processes the depth information. During the adversarial training process, the generator obtains the initial features from the graph structure model output by the GNN and generates the three-dimensional morphology layer by layer. The discriminator evaluates the performance of the generator by comparing the generated three-dimensional morphology with the real data (such as the three-dimensional morphology obtained from the actual scan). Through multiple rounds of adversarial training, the generator can gradually generate more refined three-dimensional morphology of the defect, especially in the edge details and depth information of the defect. For example, the generator can learn how to handle complex shapes during training, and identify and reconstruct tiny defects that may be ignored in other methods. In order to ensure that the generated 3D morphology is consistent with the actual operating conditions, real-time high-voltage cable operating parameters are introduced into the training of GAN. These parameters help the generator adjust the generation process of the 3D morphology so that the generated defect model is not only accurate in space, but also reflects the defect characteristics under the current operating state. This optimization step ensures the practicality and accuracy of the model.

[0081] Based on the optimized 3D morphology and combined with real-time high-voltage cable operating parameters, a time-dynamic defect distribution map is generated.

[0082] S5: inputting the defect distribution graph into a Markov random field model, and generating an optimal repair strategy through probabilistic reasoning and optimization of the graph structure;

[0083] Wherein, the S5 specifically includes the following steps:

[0084] S51: discretize the defect distribution map data and construct a state space model including different areas of the high-voltage cable lead sealing layer;

[0085] It should be noted that the defect distribution map is usually continuous three-dimensional data. In order to facilitate subsequent mathematical modeling and calculation, it needs to be discretized. The discretization process involves dividing the entire space of the lead sealing layer into several finite regions or voxels, each of which represents a state. These voxels can be adaptively adjusted according to the distribution, shape and size of the defects to ensure that important defect areas have higher resolution. Based on the discretization of data, a state space model is constructed. The state space model defines the state of each discrete area in the lead sealing layer of the high-voltage cable, and each state corresponds to the defect information of the area, including the location, depth, size, material properties, etc. of the defect. Each state in the state space model can be represented as a vector, which contains all the characteristic parameters related to the area. The model also needs to define possible transfer paths between states, which represent possible state changes between regions during the repair process, such as the state transition from defective to repaired.

[0086] S52: Based on the state space model, the Markov random field model is used to map the spatial position, depth and correlation of the defect with the adjacent area into random variables, and a state transition matrix is ​​constructed to define the transition probability between states;

[0087] It should be noted that the Markov random field (MRF) model is constructed using the discrete states in the state space model. The MRF model is a graph model in which nodes represent discrete states and edges represent the mutual influence relationship between these states. In this model, each node represents a regional state of the high-voltage cable lead sealing layer, and each edge represents the spatial correlation between different states. Through the MRF model, the spatial position, depth of the defect and its correlation with the adjacent area can be mapped into random variables, thereby capturing the complex spatial structure inside the lead sealing layer. Based on the MRF model, the state transition matrix is ​​defined. The state transition matrix describes the transition probabilities between different states, which represent the possibility of the system transferring to another state after a specific repair operation occurs in a certain state. For example, the repair operation can include material filling, crack plugging or electrical insulation repair. The state transition probability is determined by historical data and simulation results, and is corrected according to the actual situation. By defining the state transition matrix, the system can calculate the probability of the high-voltage cable lead sealing layer transferring from the current state to the target state under a specific repair strategy.

[0088] S53: Based on the state transfer matrix, applying optimal control theory, evaluating different repair strategies, calculating the expected repair effect under each state through a dynamic programming algorithm, and identifying the optimal repair path;

[0089] S54: Combined with the real-time operating parameters of the cable, a multi-scale hierarchical optimization algorithm is used to perform multiple rounds of iterative optimization on the optimal repair strategy, and finally a high-voltage cable lead layer defect repair plan based on the optimal control sequence is output, and the repair plan is dynamically adjusted through real-time monitoring and feedback mechanism.

[0090] Furthermore, the formula of the Markov random field model is as follows:

[0091] ,

[0092] in, represents the minimum expected repair cost under state s, where state s represents the defect state of different areas of the lead sealing layer of the high-voltage cable; a represents the repair actions that can be taken under state s, such as adjusting the pressure, temperature and other parameters in the lead sealing layer, or performing local physical repairs; It represents the set of feasible repair actions under state s. The specific actions are determined according to the defect model generated by multimodal data fusion and deep learning. represents the immediate repair cost of performing action a in state s, including material consumption and repair time; represents the discount factor, which is used to weigh the relationship between the immediate repair cost and the future repair cost, and is set according to the operating life of the cable and the maintenance plan; Indicates that after executing action a in state s, it transfers to state The probability of is determined by the state transfer matrix in the Markov random field model, which reflects the possible state after the sealing layer defect is repaired; S represents the set of all states, including all states that appear after the sealing layer defect is repaired.

[0093] S6: Dynamically adjust the high-voltage cable based on the optimal repair strategy, and perform real-time feedback and remote monitoring of defect detection through Internet of Things technology; the Internet of Things technology specifically combines edge computing and 5G communication technology, performs real-time data processing and analysis at edge nodes close to the high-voltage cable, and realizes instant feedback and remote control of defect detection results.

[0094] Specifically, the high-voltage cables are dynamically adjusted based on the optimal repair strategy. Using IoT technology, the repair strategy is sent to the edge computing nodes on site, which are responsible for real-time control of the operating status of the cables and executing corresponding adjustment measures, such as adjusting the current load and reducing the temperature.

[0095] The system uses IoT technology to monitor the entire process in real time and provide immediate feedback. This implementation includes the following:

[0096] Edge computing: Edge computing nodes near high-voltage cables are responsible for real-time processing and analysis of sensor data. Specifically, these nodes collect and process data on cable temperature, pressure, current, vibration and other parameters at high frequencies to quickly identify any possible abnormalities. Through a distributed computing architecture, edge nodes can complete complex analysis tasks locally, reduce data transmission delays and improve response speed.

[0097] 5G communication network: The edge computing nodes transmit the processed data and analysis results to the central control center in real time through the 5G communication network. The high bandwidth and low latency characteristics of the 5G network make the data transmission speed extremely fast, ensuring that the central control center can obtain the on-site situation in the first time, so as to make corresponding adjustments and decisions.

[0098] Real-time feedback mechanism: The central control center conducts a comprehensive analysis of the real-time data received and provides feedback based on the analysis results. For example, if it is detected that the adjusted current load is still insufficient to relieve the pressure in the defective area, the central control center can instruct the edge node to further reduce the current or take other protective measures. This feedback mechanism ensures that during the dynamic adjustment process, the system can always be optimized and adjusted according to the latest operating status.

[0099] Remote monitoring and control: Operators can view the operating status of high-voltage cables in real time through the remote monitoring platform and intervene in the detection and repair process. If the automation system cannot effectively respond in some special circumstances, operators can manually adjust the operating parameters of the cable or modify the repair strategy through the remote control interface. For example, in an emergency, the operator can instruct the system to immediately power off or switch to a backup cable to ensure the safety of the cable.

[0100] Logging and data storage: All data and operation logs generated during dynamic adjustment and monitoring are recorded in real time and stored in a central database. These data are not only used for current monitoring and adjustment, but also serve as an important reference for subsequent fault analysis, system optimization, and model training. Through regular data review and analysis, the system can continuously improve its fault prediction and handling capabilities.

[0101] In summary, the present invention collects the operating parameters of the cable in real time through an embedded sensor network, and combines phased array ultrasonic technology to perform multi-angle and multi-depth scanning, which can dynamically adjust the detection frequency and angle, thereby accurately capturing the defect information of the lead sealing layer. Multi-frequency and all-round scanning enables comprehensive detection coverage, ensuring that defects at different depths and positions can be accurately identified. Using the inverse scattering matrix analysis algorithm, the system can conduct an in-depth analysis of the scattering and reflection characteristics of electromagnetic waves, and accurately locate the three-dimensional coordinates of the defects by calculating the scattering coefficient and identifying multiple scattering paths. This multi-dimensional analysis can reconstruct the complex defect morphology inside the lead sealing layer and provide a more intuitive and comprehensive defect feature map.

[0102] Through multimodal data fusion and deep learning algorithms, the system can automatically generate and finely optimize the defect model, especially by capturing the complex spatial relationship of defects through graph neural networks, and optimizing the three-dimensional morphology of defects using generative adversarial networks, thereby improving the accuracy and operability of the defect model. By inputting the defect distribution map into the Markov random field model and combining it with optimal control theory, the system can dynamically generate the optimal repair strategy. The use of Internet of Things technology for real-time feedback and remote monitoring ensures the intelligence and efficiency of the repair process.

[0103] The present invention not only has high precision and high coverage in defect detection, but also provides a full-process integrated solution from defect detection, feature extraction to repair strategy generation and application by combining deep learning with optimized analysis of graph models, greatly improving the operational safety and reliability of the high-voltage cable lead sealing layer.

[0104] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for detecting defects in a high-voltage cable lead sealing layer, characterized in that: The following steps are involved: S1: Based on the real-time operating status of the high-voltage cable, real-time parameters of the high-voltage cable are collected through an embedded sensor network, wherein the real-time parameters include temperature, pressure, current and voltage of the high-voltage cable; S2: Use phased array ultrasonic technology to scan the lead sealing layer of the high-voltage cable, dynamically adjust the frequency and angle, and extract the key frequency domain features of the ultrasonic signal; S3: Based on key frequency domain characteristics and real-time parameters, the inverse scattering matrix analysis algorithm is used to analyze the scattering and reflection characteristics of electromagnetic waves inside the lead sealing layer, reconstruct the three-dimensional morphology of the defect, and form a complete defect feature through multi-modal data fusion; S4: Combine multimodal fusion data and use deep learning algorithms to analyze data and generate defect models, accurately locate the position, depth and size of defects, and form a complete defect distribution map; S5: inputting the defect distribution graph into a Markov random field model, and generating an optimal repair strategy through probabilistic reasoning and optimization of the graph structure; S6: Dynamically adjust the high-voltage cables based on the optimal repair strategy, and use IoT technology to provide real-time feedback and remote monitoring of defect detection; Wherein, the S3 specifically includes the following steps: S31: Based on the key frequency domain features, an inverse scattering matrix is ​​constructed and the matrix is ​​subjected to bandpass filtering; S32: Using the inverse scattering matrix analysis algorithm, the scattering coefficient of each frequency band signal is calculated, the multiple scattering paths in the cable lead sealing layer are identified, and the preliminary location of the defect is determined; S33: Time-reversing the multiple scattering paths, reconstructing the scattering scene inside the lead sealing layer through the back propagation algorithm, and accurately locating the three-dimensional coordinates of the defects; S34: matching and verifying the back propagation results of the multiple scattering paths with the key frequency domain features, and generating a three-dimensional defect feature map containing position and morphology information; The formula of the inverse scattering matrix analysis algorithm is as follows: ， in, It represents the defect distribution function from the i-th ultrasonic transmitter to the j-th receiver at frequency f, that is, the defect characteristics inside the lead sealing layer; It represents the amplitude factor of the kth scattering path, reflecting the attenuation and energy loss on the path in the lead sealing layer of the high-voltage cable; represents the time delay of the kth scattering path, the signal propagation time delay affected by the real-time parameters of the high-voltage cable, the real-time parameters including temperature, pressure, current and voltage; represents the propagation distance of the kth scattering path, which is obtained by scanning the lead sealing layer at multiple angles and depths using multi-frequency ultrasonic waves; represents the scattering attenuation index; represents the phase matrix element in the transfer matrix at frequency f; represents the noise matrix element at frequency f.

2. A method for detecting defects in a high-voltage cable lead sealing layer according to claim 1, characterized in that: The phased array ultrasonic technology includes using a multi-frequency ultrasonic transmitting array and a receiving array to control ultrasonic waves of different frequencies to scan the lead sealing layer at multiple angles and depths.

3. A method for detecting defects in a high-voltage cable lead sealing layer according to claim 1, characterized in that: The S4 specifically comprises the following steps: S41: Based on the extracted multimodal fusion data, an autoencoder is used to reduce the dimension and extract features of the defect data, and the potential features of the defects are preliminarily identified; S42: Use graph neural networks to model the graph structure of defect features, capture the complex spatial relationship of defects, and combine generative adversarial networks to perform adversarial optimization on the three-dimensional morphology of defects, and refine the defect edges and depths through adversarial training; S43: Generate a final defect distribution map based on the optimized three-dimensional morphology reconstruction.

4. A method for detecting defects in a high-voltage cable lead sealing layer according to claim 3, characterized in that: The S42 specifically includes the following steps: A dynamic graph structure model based on multimodal data is constructed. The defect characteristics of the lead sealing layer of the high-voltage cable are used as nodes in the graph structure. The edges between the nodes represent the potential causal relationship, and the edge weights are adjusted to adapt to changes in the operating environment. Adopting adaptive graph neural network, it captures the complex spatial characteristics of defects through self-supervised learning, identifies new defects through topological learning, and updates the graph structure in real time; Generative adversarial networks are introduced to use multi-channel parallel computing to model the three-dimensional morphology of defects layer by layer; Based on the optimized 3D morphology and combined with real-time high-voltage cable operating parameters, a time-dynamic defect distribution map is generated.

5. A method for detecting defects in a high-voltage cable lead sealing layer according to claim 1, characterized in that: The S5 specifically includes the following steps: S51: discretize the defect distribution map data and construct a state space model including different areas of the high-voltage cable lead sealing layer; S52: Based on the state space model, the Markov random field model is used to map the spatial position, depth and correlation of the defect with the adjacent area into random variables, and a state transition matrix is ​​constructed to define the transition probability between states; S53: Based on the state transfer matrix, applying optimal control theory, evaluating different repair strategies, calculating the expected repair effect under each state through a dynamic programming algorithm, and identifying the optimal repair path; S54: Combined with the real-time operating parameters of the cable, a multi-scale hierarchical optimization algorithm is used to perform multiple rounds of iterative optimization on the optimal repair strategy, and finally a high-voltage cable lead layer defect repair plan based on the optimal control sequence is output, and the repair plan is dynamically adjusted through real-time monitoring and feedback mechanism.

6. A method for detecting defects in a high-voltage cable lead sealing layer according to claim 5, characterized in that: The formula of the Markov random field model is as follows: ， in, represents the minimum expected repair cost under state s, where state s represents the defect state of different areas of the lead sealing layer of the high-voltage cable; a represents the repair action that can be taken under state s; represents the set of feasible repair actions in state s; represents the immediate repair cost of performing action a in state s, including material consumption and repair time; represents the discount factor, which is used to weigh the relationship between the immediate repair cost and the future repair cost, and is set according to the operating life of the cable and the maintenance plan; Indicates that after executing action a in state s, it transfers to state The probability of; S represents the set of all states, including all states that appear after the sealing layer defect is repaired.

7. A method for detecting defects in a high-voltage cable lead sealing layer according to claim 1, characterized in that: The Internet of Things technology specifically combines edge computing and 5G communication technology to perform real-time data processing and analysis at edge nodes close to high-voltage cables, and achieve instant feedback and remote control of defect detection results.

Images