An on-line monitoring method and device for the insulation performance of power cables

Abstract

The present invention proposes an on-line monitoring method and device for the insulation performance of power cables. The method includes: constructing a multi-modal sensor network to synchronously collect multi-dimensional cable data, establishing an electromagnetic-thermal-chemical multi-physical field coupling model, and real-time analyzing the insulation state of the cable. A hybrid deep learning model is used to monitor insulation defects and predict the remaining service life. Based on spatio-temporal characteristics, a digital twin method is adopted to dynamically generate warning thresholds and adaptively adjust the monitoring strategy. The present invention breaks through the limitation of single-parameter monitoring by synchronously collecting multi-dimensional data through a multi-modal sensor network. The multi-dimensional data fusion accurately captures the insulation degradation characteristics of the cable, and the constructed coupling model realizes the quantitative description of energy transfer and material evolution. Through the spatial encoder of the residual network and dilated convolution, combined with the bidirectional LSTM-GRU time encoder, the spatio-temporal characteristics of the cable are extracted. The cross-modal attention mechanism fuses multi-source features to improve the accuracy of defect identification and life prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of cable insulation performance monitoring, and particularly to an on-line monitoring method and device for the insulation performance of power cables. Background Art

[0002] As a core component of the power grid transmission and distribution system, the insulation performance of power cables directly affects the safe operation of the power system. With the increase in grid load density and the prominent problem of cable aging, the traditional off-line detection mode is difficult to meet the requirements of real-time, accuracy, and full-domain coverage.

[0003] Traditional monitoring methods mostly rely on a single type of sensor (such as only temperature or partial discharge detection), and cannot synchronously obtain multi-dimensional characteristics (electromagnetic, thermal, chemical, etc.) of the cable insulation state. For example, there is a strong spatio-temporal correlation between partial discharge pulses and temperature field distribution, but due to the lack of multi-modal data fusion ability in existing systems, the accuracy of defect diagnosis is insufficient. Existing cable condition assessment models are mostly based on a single physical field (such as heat conduction model), ignoring the electromagnetic-thermal-chemical multi-field coupling effect. The essence of cable insulation deterioration is the result of the synergistic action of electro-thermal-chemical multi-physical fields, and traditional simplified models are difficult to accurately describe the defect evolution mechanism (such as the chain reaction of temperature gradient and chemical degradation caused by partial discharge). The warning thresholds of existing monitoring systems mostly adopt fixed standards (such as the temperature limit specified by IEC), without considering the dynamic characteristics of cable operating conditions (such as load fluctuations, sudden changes in ambient temperature) and the real-time impact of defect evolution rate on the threshold, resulting in high false alarm / missed alarm rates. Traditional monitoring strategies adopt fixed sampling frequencies and sensor working modes, and cannot adaptively adjust according to the cable health status. For example, high-frequency monitoring is still used in the early stage of defect development, resulting in waste of resources; while low-frequency sampling is still maintained in critical states, delaying the timing of fault handling. Summary of the Invention

[0004] The present invention aims to at least solve the technical problem of insufficient accuracy of defect diagnosis in the prior art, and particularly innovatively provides an on-line monitoring method and device for the insulation performance of power cables.

[0005] In order to achieve the above object of the present invention, the present invention provides an on-line monitoring method for the insulation performance of power cables, the method comprising:

[0006] S1. Construct a multi-modal sensor network, and synchronously collect multi-dimensional cable data based on the multi-modal sensor network, the multi-dimensional cable data including partial discharge pulse data, distributed temperature field data, dielectric loss factor, and chemical characteristic gas data;

[0007] S2. Establish an electromagnetic-thermal-chemical multi-physical field coupling model, input the multi-dimensional cable data into the electromagnetic-thermal-chemical multi-physical field coupling model, and analyze the multi-dimensional evolution law of the cable insulation state in real time;

[0008] S3. Establish a hybrid deep learning model, use the hybrid deep learning model to extract the spatio-temporal features in the multi-dimensional evolution law, monitor the insulation defects of the cable based on the spatio-temporal features, and predict the remaining service life of the cable;

[0009] S4. Based on the spatio-temporal features, use the digital twin method to dynamically generate warning thresholds, and adaptively adjust the monitoring strategy based on the warning thresholds and working condition parameters.

[0010] As an optional embodiment of the present invention, optionally, in step S1, synchronously collecting multi-dimensional cable data based on the multi-modal sensor network includes:

[0011] S101. Use a partial discharge detection system combining a differential Rogowski coil array and a high-frequency current transformer to collect partial discharge pulse data;

[0012] S102. Use a distributed optical fiber temperature measurement system based on the Raman scattering principle to collect distributed temperature field data;

[0013] S103. Use a dielectric loss factor monitoring module including a programmable frequency source and an orthogonal sampling phase-sensitive demodulation circuit to collect the dielectric loss factor;

[0014] S104. Use a gas monitoring unit composed of a nanotube array gas sensor and an optical fiber evanescent wave sensor to collect chemical characteristic gas data.

[0015] As an optional embodiment of the present invention, optionally, in step S2, establishing an electromagnetic-thermal-chemical multi-physical field coupling model includes:

[0016] S201. Use the partial discharge pulse data to establish an electromagnetic field model;

[0017] S202. Use the distributed temperature field data and the dielectric loss factor to establish a heat conduction model;

[0018] S203. Use the chemical characteristic gas data to establish a chemical reaction field model;

[0019] S204. Couple the electromagnetic field model with the heat conduction model, and couple the heat conduction model with the chemical reaction field model to obtain an electromagnetic-thermal-chemical multi-physical field coupling model;

[0020] S205. Evaluate the confidence level of the electromagnetic-thermal-chemical multi-physical field coupling model, and optimize the electromagnetic-thermal-chemical multi-physical field coupling model based on the evaluation results.

[0021] As an alternative embodiment of the present invention, optionally, establishing the hybrid deep learning model in step S3 includes:

[0022] S301. Perform spatio-temporal alignment and normalization processing on the multi-dimensional cable data to construct a tensor input including space-time dimensions;

[0023] S302. Construct a spatial encoder based on the combination of a residual network and dilated convolution, and use the spatial encoder to extract spatial feature vectors from the tensor input;

[0024] S303. Cascade a bidirectional long short-term memory network and a gated recurrent unit to obtain a time encoder, and use the time encoder to extract the temporal feature vectors of the insulation performance from the tensor input;

[0025] S304. Dynamically weight and fuse the spatial feature vectors and the temporal feature vectors based on a cross-modal attention fusion module to obtain spatio-temporal features;

[0026] S305. Construct a cable structure life prediction model based on the spatio-temporal features using a graph convolutional network;

[0027] S306. Construct a cable insulation defect identification model based on the spatio-temporal features using a lightweight classifier, and identify the types of cable insulation defects based on the cable insulation defect identification model;

[0028] S307. Train and validate the cable structure life prediction model and the cable insulation defect identification model respectively;

[0029] S308. Integrate the trained cable structure life prediction model and the cable insulation defect identification model to obtain a hybrid deep learning model.

[0030] As an alternative embodiment of the present invention, optionally, adaptively adjusting the monitoring strategy based on the warning threshold and operating conditions parameters in step S4 includes:

[0031] S401. Construct a digital twin model based on the multi-dimensional evolution law of the cable insulation state;

[0032] S402. The digital twin model dynamically generates a warning threshold for the cable insulation performance based on the currently collected multi-dimensional cable data and the operating conditions parameters of the cable using a machine learning method;

[0033] S403. When the insulation defect of the monitored cable approaches the warning threshold of the cable insulation performance, trigger a warning mechanism;

[0034] S404. If the insulation defect of the monitored cable is far from the warning threshold, reduce the monitoring frequency;

[0035] S405. If the insulation defect of the monitoring cable approaches or exceeds the warning threshold, increase the monitoring frequency;

[0036] S406. Evaluate the effect of the adjusted monitoring strategy. If the evaluation result shows that the adjusted monitoring strategy fails to effectively improve the monitoring accuracy, then re-execute steps S401 to S405.

[0037] On the other hand, the present invention also provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the on-line monitoring method for the insulation performance of the power cable.

[0038] Advantages of the present invention: Through the multi-modal sensor network, the present invention realizes the synchronous acquisition of partial discharge pulses, temperature fields, dielectric losses, and chemical gases, breaking through the limitations of traditional single-parameter monitoring. The multi-dimensional data fusion can accurately capture the multi-physical characteristics of cable insulation degradation (such as the spatio-temporal correlation between the temperature gradient caused by partial discharge and chemical degradation). The constructed electromagnetic-thermal-chemical coupling model correlates electromagnetic energy consumption, heat conduction, and chemical reaction kinetics through the Hamiltonian operator to realize the quantitative description of energy transfer and material evolution during the cable insulation degradation process. It can accurately analyze complex degradation mechanisms such as water tree aging and local carbonization. By adopting a spatial encoder composed of a residual network and a dilated convolution, combined with a bidirectional LSTM-GRU cascaded time encoder, the spatio-temporal correlation features of the cable state tensor are effectively extracted. The cross-modal attention mechanism dynamically weights and fuses multi-source features, improving the probability of insulation defect recognition and the accuracy of remaining life prediction. The dynamically generated dynamic warning threshold based on the digital twin model can adapt to changes in working condition parameters (such as load rate, ambient temperature, etc.), reducing the false alarm rate. The closed-loop optimization mechanism of the monitoring strategy (the sampling frequency is increased by 3-5 times when the defect approaches the threshold, and the frequency is automatically reduced during the stable period) reduces the monitoring energy consumption and significantly improves the operation and maintenance economy.

[0039] The additional aspects and advantages of the present invention will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present invention. Description of the Drawings

[0040] The above and / or additional aspects and advantages of the present invention will become obvious and easy to understand from the description of the embodiments in conjunction with the following drawings, where:

[0041] Figure 1 is a flowchart of an on-line monitoring method for the insulation performance of a power cable according to Embodiment 1 of the present invention;

[0042] Figure 2 is a schematic structural diagram of an electronic device according to Embodiment 2 of the present invention. Detailed Embodiments

[0043] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0044] Example 1

[0045] like Figure 1 As shown, a method for online monitoring of insulation performance of a power cable, the method comprising:

[0046] S1. Building a multimodal sensor network, and synchronously collecting multidimensional cable data based on the multimodal sensor network, wherein the multidimensional cable data includes partial discharge pulse data, distributed temperature field data, dielectric loss factor, and chemical characteristic gas data;

[0047] It should be noted that in step S1, the multimodal sensor network is a multimodal sensor network composed of differential Rogowski coils, distributed optical fiber temperature measurement, dielectric loss monitoring and nanotube gas sensors. Specifically, the differential Rogowski coil array is combined with a high-frequency current transformer to achieve high-sensitivity capture of local discharge pulses and ensure accurate acquisition of local discharge signals. The distributed optical fiber temperature measurement system is based on the principle of Raman scattering, distributed along the length of the cable, and monitors the temperature field distribution in real time to provide detailed temperature data. The dielectric loss factor monitoring module integrates a programmable frequency source and an orthogonal sampling phase-sensitive demodulation circuit to accurately measure the dielectric loss of the cable and reflect the loss characteristics of the cable insulation material. The gas monitoring unit composed of a nanotube array gas sensor and an optical fiber evanescent wave sensor can sensitively detect chemical characteristic gases released inside the cable, such as carbon monoxide. These gases are products of thermal decomposition or chemical degradation of cable insulation materials and are of great value for evaluating the insulation status of the cable.

[0048] S2. Establishing an electromagnetic-thermal-chemical multi-physics field coupling model, inputting the multi-dimensional cable data into the electromagnetic-thermal-chemical multi-physics field coupling model, and analyzing the multi-dimensional evolution law of the cable insulation state in real time;

[0049] It should be noted that in this embodiment, an electromagnetic-thermal-chemical multi-physics field coupling model is established, and the local discharge pulse data is mapped into an electromagnetic field model, a temperature field and dielectric loss driven heat conduction model, and a chemical gas concentration triggering a chemical reaction field model, and the bidirectional coupling relationship between the three fields is described by partial differential equations (such as Joule heat driving the temperature field, and the chemical reaction rate is affected by temperature). The model combines the control equations of electromagnetic, thermal, and chemical fields through the Hamiltonian operator, introduces a confidence assessment mechanism (such as residual analysis) to dynamically optimize the model parameters, and improves the analytical accuracy of the insulation degradation chain reaction.

[0050] S3. Establish a hybrid deep learning model, use the hybrid deep learning model to extract spatio-temporal features in the multi-dimensional evolution law, monitor the insulation defects of the cable based on the spatio-temporal features, and predict the remaining service life of the cable;

[0051] It should be noted that the hybrid deep learning model adopts a dual-encoder architecture:

[0052] Spatial encoder: Combine the residual network and dilated convolution to extract the spatial distribution features of insulation defects (such as the morphology of partial discharge clusters);

[0053] Temporal encoder: Cascade bidirectional LSTM and GRU to capture the temporal law of insulation performance degradation (such as the growth trend of dielectric loss).

[0054] The cross-modal attention mechanism dynamically fuses spatio-temporal features. The graph convolutional network (GCN) combines the cable topology structure to predict the remaining life, and the lightweight classifier (such as MobileNet) real-time identifies the defect types (such as water trees, electrical erosion holes).

[0055] S4. Based on the spatio-temporal features, use the digital twin method to dynamically generate warning thresholds, and adaptively adjust the monitoring strategy based on the warning thresholds and operating condition parameters.

[0056] It should be noted that based on the digital twin model, the physical state of the cable is mapped in real time. Use historical data and online learning (such as incremental SVM) to dynamically generate warning thresholds sensitive to operating conditions, and adjust the monitoring strategy according to the defect development rate (derived from spatio-temporal features):

[0057] Far from the threshold: Reduce the sampling frequency (such as from 1 kHz to 100 Hz);

[0058] Close to the threshold: Increase the monitoring density and trigger local enhanced scanning (such as focusing on the suspected defect area).

[0059] Through closed-loop effect evaluation (such as comprehensive optimization of F1-score and computing resource consumption), iteratively update the digital twin model and the monitoring strategy to achieve a balance between resource efficiency and diagnostic accuracy.

[0060] In summary, the online monitoring method for the insulation performance of the power cable in this embodiment realizes the synchronous acquisition of multi-dimensional characteristics of the cable insulation state by constructing a multi-modal sensor network, effectively solving the problems of single data and insufficient accuracy of defect diagnosis in the prior art. This method not only breaks through the limitations of traditional monitoring technologies, but also realizes the quantitative description of energy transfer and material evolution during the cable insulation degradation process through the establishment of an electromagnetic-thermal-chemical multi-physical field coupling model, so as to accurately analyze complex degradation mechanisms.

[0061] In addition, the application of the hybrid deep learning model further improves the probability of insulation defect identification and the accuracy of remaining life prediction. Through the dual-encoder architecture, this model effectively extracts the spatio-temporal correlation features of the cable state tensor and realizes the dynamic weighted fusion of multi-source features through the cross-modal attention mechanism. This not only improves the accuracy of monitoring.

[0062] Finally, the warning threshold dynamically generated based on the digital twin method enables the monitoring strategy to be adaptively adjusted according to the operating condition parameters of the cable and the defect development rate. This not only reduces the false alarm rate, but also continuously optimizes the digital twin model and the monitoring strategy through the closed-loop effect evaluation mechanism, achieving a balance between resource efficiency and diagnostic accuracy.

[0063] As an optional embodiment of the present invention, optionally, the synchronous acquisition of multi-dimensional cable data based on the multi-modal sensor network in step S1 includes:

[0064] S101. Collect partial discharge pulse data using a partial discharge detection system that combines a differential Rogowski coil array and a high-frequency current transformer;

[0065] It should be noted that in step S101, the combination of the differential Rogowski coil array and the high-frequency current transformer realizes the high-precision capture of the cable partial discharge signal. The differential Rogowski coil array has the characteristics of high sensitivity and fast response speed, and can accurately capture the weak pulse signals generated by partial discharge. The high-frequency current transformer can amplify and convert these pulse signals for subsequent data processing and analysis.

[0066] S102. Collect distributed temperature field data using a distributed optical fiber temperature measurement system based on the Raman scattering principle;

[0067] It should be noted that in step S102, the distributed optical fiber temperature measurement system based on the Raman scattering principle can perform continuous temperature measurement along the cable length and provide high-precision temperature distribution data. This system uses optical fiber as the sensing element and measures the temperature by detecting the intensity change of Raman scattered light. It has the advantages of wide measurement range, high precision, and immunity to electromagnetic interference. The collected distributed temperature field data can reflect the temperature state of the cable at different positions.

[0068] S103. Collect the dielectric loss factor using a dielectric loss factor monitoring module that includes a programmable frequency source and an orthogonal sampling phase-sensitive demodulation circuit;

[0069] It should be noted that in step S103, the design of the dielectric loss factor monitoring module adopts advanced electronic measurement technology to ensure the accurate measurement of the dielectric loss factor. The programmable frequency source can output signals of different frequencies according to needs to meet the dielectric loss measurement requirements of different cables. The quadrature sampling phase-sensitive demodulation circuit can efficiently extract the dielectric loss signal, eliminate interference, and improve the measurement accuracy. The dielectric loss factor data collected by this module can reflect the loss degree of the cable insulation material.

[0070] S104. Use the gas monitoring unit composed of the nanotube array gas sensor and the fiber optic evanescent wave sensor to collect chemical characteristic gas data.

[0071] It should be noted that in step S104, the nanotube array gas sensor has the characteristics of high sensitivity and fast response, and can accurately detect trace chemical characteristic gases released inside the cable. The fiber optic evanescent wave sensor uses the evanescent wave effect of light to measure the gas concentration with high precision. The combined use of these two sensors improves the accuracy and reliability of the detection of chemical characteristic gases. The collected chemical characteristic gas data can reflect the degree of thermal decomposition or chemical degradation of the cable insulation material.

[0072] As an alternative embodiment of the present invention, optionally, establishing the electromagnetic-thermal-chemical multi-physical field coupling model in step S2 includes:

[0073] S201. Use the partial discharge pulse data to establish an electromagnetic field model;

[0074] S202. Use the distributed temperature field data and the dielectric loss factor to establish a heat conduction model;

[0075] S203. Use the chemical characteristic gas data to establish a chemical reaction field model;

[0076] S204. Couple the electromagnetic field model with the heat conduction model, and couple the heat conduction model with the chemical reaction field model to obtain an electromagnetic-thermal-chemical multi-physical field coupling model;

[0077] S205. Evaluate the confidence level of the electromagnetic-thermal-chemical multi-physical field coupling model, and optimize the electromagnetic-thermal-chemical multi-physical field coupling model based on the evaluation results.

[0078] It should be noted that in step S205, the confidence evaluation mechanism is carried out by comparing the consistency between the model prediction results and experimental data or historical data. Specifically, methods such as residual analysis, correlation coefficient calculation, or cross-validation can be used to evaluate the confidence of the model. If the evaluation results show that there is a large deviation between the model prediction results and the actual situation, it is necessary to adjust the model parameters or optimize the model structure to improve the accuracy and reliability of the model. Through continuous iteration and optimization, an electromagnetic-thermal-chemical multi-physical field coupling model that can accurately reflect the multi-dimensional evolution law of the cable insulation state is finally obtained.

[0079] As an alternative embodiment of the present invention, optionally, the expression of the electromagnetic-thermal-chemical multi-physical field coupling model is:

[0080] Electromagnetic field model:

[0081] ;

[0082] The purpose of constructing the electromagnetic field model is to describe the electromagnetic field distribution generated by partial discharge in the cable insulation and its influence on materials. The electromagnetic field model is established through the magnetic vector potential equation and the current continuity equation. Specifically, after the partial discharge pulse is collected by the high-frequency current transformer and the differential Rogowski coil array, it is input into the model as the excitation source ( ); the equation indirectly calculates the electric field strength ( ) and the magnetic field strength ( ) by solving the distribution of the magnetic vector potential ); the electromagnetic field energy is coupled to the heat conduction model through the Joule heat term ( ). When constructing the electromagnetic field model, the finite element method (FEM) or the finite difference time domain method (FDTD) is used to discretize the solution region, and the boundary conditions (such as the perfect electric conductor boundary) are set in combination with the cable geometric structure (such as the insulation layer, the shielding layer).

[0083] Heat conduction model:

[0084] ;

[0085] The purpose of constructing the heat conduction model is to describe the cable temperature field distribution and the influence of thermal stress on the insulation material. The heat conduction model is established based on Fourier's law and the energy conservation equation. The distributed optical fiber temperature measurement system provides the initial temperature field data ( ), the dielectric loss factor is collected by the orthogonal sampling phase-sensitive demodulation circuit and used to calculate the heat generated by the dielectric loss; the Joule heat term ( ) is input in real time by the electromagnetic field model to form the electro-thermal coupling; the temperature gradient drives the heat conduction process, and the high-temperature region accelerates the aging of the insulation material. The finite volume method (FVM) or the finite element method is used to solve the transient temperature field when establishing the heat conduction model; the nonlinear thermal conductivity is introduced ( to reflect the temperature dependence of the material.

[0086] Chemical reaction field model:

[0087] ;

[0088] The chemical reaction field model is established to describe the diffusion and reaction kinetics of chemical gases generated by the aging of insulating materials. The chemical reaction field model is established by combining Fick's diffusion law and Arrhenius reaction equation; the nanotube array gas sensor and the fiber optic evanescent wave sensor collect the chemical gas concentration in real time ( ), which is used as the model input; the temperature field ( ) regulates the reaction rate ( ) through the Arrhenius equation, and high temperature accelerates chemical degradation; the diffusion term ( ) describes the concentration gradient-driven diffusion of gas in the insulating layer. The finite difference method (FDM) is used to solve the reaction-diffusion equation for the chemical reaction field model; the temperature output of the coupled heat conduction model is used to dynamically update the reaction rate constant.

[0089] Among them, represents the Hamiltonian operator, represents the magnetic permeability of the material, represents the magnetic vector potential, represents the conductivity, represents the partial derivative of the magnetic vector potential with respect to time, represents the source current density, represents the material density, represents the specific heat capacity, represents the temperature, represents the thermal conductivity, represents the Joule heat term, represents the concentration of chemical products, represents the diffusion coefficient, represents the frequency factor, represents the activation energy, represents the gas constant, represents the reaction kinetics function.

[0090] Electromagnetic-thermal coupling: The Joule heat term ( ) of the electromagnetic field model is used as the heat source to input the heat conduction model;

[0091] Thermal-chemical coupling: The temperature field ( ) of the heat conduction model regulates the chemical reaction rate ( );

[0092] Bidirectional feedback: Chemical degradation changes the conductivity ( ) and thermal conductivity ( ), thereby affecting the distribution of the electromagnetic field and the temperature field, and forming a closed-loop coupling.

[0093] As an optional embodiment of the present invention, optionally, establishing the hybrid deep learning model in step S3 includes:

[0094] S301. Perform spatio-temporal alignment and normalization processing on the multi-dimensional cable data to construct a tensor input including spatial-temporal dimensions;

[0095] It should be noted that in step S301, spatio-temporal alignment is to ensure that data from different sensors can be aligned in the same time frame and spatial position for subsequent feature extraction and pattern recognition. Normalization processing is to eliminate the dimensional differences between different data dimensions and improve the training efficiency and accuracy of the model. By constructing a tensor input including spatial-temporal dimensions, the hybrid deep learning model can more effectively capture the multi-dimensional evolution law of the cable insulation state.

[0096] S302. Construct a spatial encoder based on the combination of a residual network and dilated convolution, and use the spatial encoder to extract spatial feature vectors in the tensor input;

[0097] It should be noted that in step S302, the spatial encoder uses a residual network (ResNet) as the basic architecture, and introduces skip connections to alleviate the vanishing gradient problem in deep networks, thereby improving the training stability and feature extraction ability of the model. The introduction of dilated convolution can increase the receptive field of the convolutional kernel, enabling the model to consider more extensive context information when capturing spatial features. Through the action of the spatial encoder, the key spatial features in the tensor input are effectively extracted and encoded as spatial feature vectors.

[0098] S303. Cascade a bidirectional long short-term memory network and a gated recurrent unit to obtain a time encoder, and use the time encoder to extract temporal feature vectors of the insulation performance in the tensor input;

[0099] It should be noted that in step S303, the time encoder combines the advantages of the bidirectional long short-term memory network (Bi-LSTM) and the gated recurrent unit (GRU). Bi-LSTM can process sequence data simultaneously from two directions (forward and backward), capture context information, and enhance the ability to model temporal dependencies. GRU simplifies the structure of LSTM by introducing update gates and reset gates, improving computational efficiency while maintaining good feature extraction performance. Cascading Bi-LSTM and GRU not only retains detailed temporal information but also improves the running efficiency of the model. The time encoder analyzes the time series in the tensor input, extracts the features of the insulation performance changing over time, and encodes them into temporal feature vectors.

[0100] S304. Dynamically weighted fusion of the spatial feature vector and the temporal feature vector based on the cross-modal attention fusion module to obtain spatio-temporal features;

[0101] It should be noted that in step S304, the cross-modal attention fusion module first calculates the attention weights between the spatial feature vector and the temporal feature vector, and these weights reflect the relative importance of different features in describing the cable insulation state. Then, based on these weights, dynamic weighted fusion of the spatial feature vector and the temporal feature vector is performed to generate a spatio-temporal feature vector that combines spatial and temporal information. This process not only enhances the model's ability to capture the multi-dimensional evolution law of the cable insulation state but also improves the model's recognition accuracy for different working conditions and defect types.

[0102] S305. Construct a cable structure life prediction model based on the spatio-temporal features using a graph convolutional network;

[0103] It should be noted that in step S305, a graph convolutional network (GCN) is used to construct a cable structure life prediction model because GCN can efficiently process graph-structured data and capture complex relationships between nodes. In the graph convolutional network, each part of the cable (such as the insulation layer, conductor, etc.) is represented as a node in the graph, and their connection relationships (such as physical connections, heat conduction paths, etc.) are represented as edges in the graph. Through graph convolution operations, the model can learn the structural information contained in these nodes and edges, and then accurately predict the life of the cable. Specifically, the spatio-temporal feature vector is used as the input of the graph convolutional network. After multiple layers of graph convolution operations, the feature representation of each node is obtained. Then, using these feature representations and the prior knowledge of the cable structure, a cable structure life prediction model is constructed. This model can comprehensively consider multiple aspects of information such as the electromagnetic, thermal, and chemical aspects of the cable, as well as the interaction relationships between them, to achieve accurate prediction of the cable life.

[0104] S306. Construct a cable insulation defect recognition model using a lightweight classifier based on the spatio-temporal features, and identify the types of cable insulation defects based on the cable insulation defect recognition model;

[0105] It should be noted that in step S306, algorithms such as convolutional neural network (CNN) or support vector machine (SVM) are adopted for the lightweight classifier. These algorithms have low computational complexity and memory occupancy while maintaining high recognition accuracy. The spatio-temporal feature vector is used as the input of the lightweight classifier. After feature extraction and classification decision-making, the type of cable insulation defect is output. To further improve the generalization ability and robustness of the model, data augmentation techniques (such as adding random noise, image rotation, etc.) can be used to preprocess the training data to increase the diversity of the model's training samples.

[0106] S307. Train and validate the cable structure life prediction model and the cable insulation defect recognition model respectively;

[0107] It should be noted that in step S307, the training process is carried out using a synthetic data set with labels and an experimental data set. The labels include the actual life of the cable and the known types of insulation defects. The synthetic data set is generated by simulating the operating conditions of the cable under different working conditions, including changes in various electromagnetic, thermal, and chemical parameters, and their effects on the cable insulation state. The experimental data set comes from actual cable insulation performance monitoring experiments, including various defect types and different degrees of aging. In the training process, a cross-validation strategy is adopted to evaluate the performance of the model to ensure that the model can show good generalization ability on different data sets. The validation process is carried out by comparing the consistency between the model prediction results and the experimental verification data to verify the accuracy and reliability of the model. Through training and validation, a hybrid deep learning model that can accurately predict the cable life and identify the types of insulation defects is finally obtained.

[0108] S308. Integrate the trained cable structure life prediction model and the cable insulation defect recognition model to obtain a hybrid deep learning model.

[0109] It should be noted that in step S308, integrating the cable structure life prediction model and the cable insulation defect recognition model is achieved by constructing a unified model framework. This framework can simultaneously receive the spatio-temporal feature vector as the input and call the two sub-models for prediction respectively. Specifically, for the cable structure life prediction, the framework passes the spatio-temporal feature vector to the graph convolutional network part. After multiple layers of graph convolutional operations, the predicted life of the cable is output. For the cable insulation defect recognition, the framework passes the spatio-temporal feature vector to the lightweight classifier part. After feature extraction and classification decision-making, the type of insulation defect is output.

[0110] As an alternative embodiment of the present invention, optionally, the expression for the temporal feature vector of the insulation performance extracted from the tensor input by the temporal encoder in step S302 is:

[0111] ,

[0112] ;

[0113] wherein, represents the initial layer spatial feature map, represents the spatial distribution data output by the electromagnetic-thermal-chemical multi-physics field coupling model, represents the -th layer spatial feature map, represents the activation function, represents the -th layer dilated convolution kernel, represents the dilated convolution operation, represents the -th layer spatial feature map, represents the -th layer bias;

[0114] The expression for the temporal feature vector of the insulation performance extracted from the tensor input by the temporal encoder in step S303 is:

[0115] ,

[0116] ,

[0117] ,

[0118] ;

[0119] wherein, represents the hidden state vector of the forward LSTM at time step , represents the long short-term memory network, represents the input vector at time step , that is, the time series output of the electromagnetic-thermal-chemical multi-physics field coupling model, represents the hidden state vector of the forward LSTM at time step , represents the hidden state vector of the forward and backward LSTM at time step , represents the hidden state vector of the forward and backward LSTM at time step , represents the concatenated hidden state of the bidirectional long short-term memory network at time step ​ Indicates a splicing operation, Indicates the hidden state of the gated recurrent unit at time step of Indicates the gated recurrent unit operation, Indicates the hidden state of the gated recurrent unit at time step of

[0120] As an alternative embodiment of the present invention, optionally, the expression for the cross-modal attention fusion module to dynamically weight and fuse the spatial feature vector and the temporal feature vector in step S304 is:

[0121] ,

[0122] ,

[0123] ,

[0124] ;

[0125] wherein, represents the projected spatial feature vector, represents the learned projection matrix for mapping the spatial feature vector and the temporal feature vector to a common dimension, represents global average pooling of the spatial feature map, represents the bias term of the projection layer, represents the projected temporal feature vector, the learned projection matrix for mapping the temporal feature vector and the spatial feature vector to a common dimension, represents temporal dimension pooling of the temporal feature vector, represents the bias term of the projection layer, represents the dynamically generated attention weight vector, represents the activation function, represents the transpose of the projected spatial feature vector, represents the learnable attention matrix, represents the scaling factor, represents the spatio-temporal feature, represents element-wise multiplication.

[0126] As an alternative embodiment of the present invention, optionally, the expression for monitoring the insulation defect of the cable based on the spatio-temporal feature in step S3 is:

[0127] ;

[0128] wherein, represents the predicted probability vector, Denotes an activation function, Denotes the parameters of the classification layer, Denotes spatio-temporal features, Denotes the bias vector of the classification layer;

[0129] In step S3, monitoring the cable based on the spatio-temporal features to predict the remaining service life of the cable includes:

[0130] ,

[0131] ,

[0132] ,

[0133] ,

[0134] ;

[0135] Among them, Denotes the encoded hidden state sequence, Denotes encoding the input feature matrix, Denotes the time step Of the normalized weight, Denotes the transpose of the feature transformation matrix, Denotes the feature transformation matrix, Denotes the time step Of the hidden state sequence, Denotes the bias term for attention calculation, Denotes the time step Of the hidden state sequence, Denotes the context vector, Denotes the predicted original life value, Denotes the decoding weight matrix, Denotes the decoding bias term, Denotes the life prediction value, Denotes the physical decay factor, Denotes the insulation aging rate coefficient, Denotes the current time step.

[0136] As an optional embodiment of the present invention, optionally, in step S4, adaptively adjusting the monitoring strategy based on the warning threshold and operating conditions parameters includes:

[0137] S401. Construct a digital twin model based on the multi-dimensional evolution law of the cable insulation state;

[0138] It should be noted that in step S401, the digital twin model realizes high-precision simulation of the cable insulation state by integrating multi-dimensional information such as the physical structure, material properties, and operating environment of the cable. This model can reflect the evolution of the cable insulation state under different working conditions in real time. The specific establishment of the digital twin model is as follows: First, collect the physical structure parameters of the cable, such as conductor size, insulation layer thickness, etc., and material property data, such as the dielectric constant and thermal conductivity of the insulation material. Then, according to the actual operating environment of the cable, including factors such as temperature, humidity, and electromagnetic field strength, set the corresponding simulation conditions. Next, use advanced simulation software to input the above information into the model and perform coupled simulation of multiple physical fields to simulate the change of the cable insulation state under different working conditions. By continuously iterating and optimizing the model parameters, the output result of the digital twin model is highly consistent with the actual situation, so as to achieve high-precision simulation and real-time reflection of the cable insulation state.

[0139] S402. The digital twin model dynamically generates a warning threshold for the cable insulation performance based on the currently collected multi-dimensional cable data and the working condition parameters of the cable by using machine learning methods;

[0140] It should be noted that in step S402, the digital twin model not only considers the physical structure and material properties of the cable, but also combines real-time operating data, such as current, voltage, temperature, etc., and historical fault records, and deeply mines and analyzes these data through machine learning algorithms, so as to dynamically adjust the warning threshold. This dynamic adjustment mechanism can more accurately reflect the current insulation state of the cable, avoid false alarms or missed alarms, and improve the accuracy and reliability of monitoring. The generation of the warning threshold is specifically as follows: First, extract the insulation performance data of the cable from the historical database, including key indicators such as insulation resistance and leakage current. Then, use machine learning algorithms, such as support vector machines and random forests, to train these data to establish a mapping relationship between the insulation performance and the warning threshold. Next, according to the currently collected multi-dimensional cable data and working condition parameters, update the input of the model in real time and dynamically generate a new warning threshold. Finally, compare the generated warning threshold with the real-time monitored insulation performance data. If it exceeds the threshold, trigger the warning mechanism and notify the operation and maintenance personnel to handle it in time.

[0141] S403. When the insulation defect of the monitored cable approaches the warning threshold of the cable insulation performance, trigger the warning mechanism;

[0142] S404. If the insulation defect of the monitored cable is far from the warning threshold, reduce the monitoring frequency;

[0143] S405. If the insulation defect of the monitored cable approaches or exceeds the warning threshold, increase the monitoring frequency;

[0144] It should be noted that in steps S404 and S405, by dynamically adjusting the monitoring strategy, refined management and efficient monitoring of the cable insulation status can be achieved. Specifically, when it is detected that the insulation defect of the cable is gradually approaching the warning threshold, it indicates that the insulation performance of the cable may be gradually deteriorating. At this time, the warning mechanism is triggered, which can remind the operation and maintenance personnel to take measures for intervention in a timely manner to prevent the occurrence of insulation faults. When it is detected that the insulation defect of the cable is far from the warning threshold, it indicates that the insulation status of the cable is relatively stable. At this time, the monitoring frequency can be appropriately reduced to reduce unnecessary monitoring costs. On the contrary, when it is detected that the insulation defect of the cable approaches or exceeds the warning threshold, it indicates that there may be serious problems with the insulation performance of the cable. At this time, the monitoring frequency needs to be increased to pay closer attention to the insulation status of the cable and ensure the safe operation of the cable. In this way, refined management and efficient monitoring of the cable insulation status can be achieved, improving the operation safety and reliability of the cable.

[0145] S406. Evaluate the effect of the adjusted monitoring strategy. If the evaluation result shows that the adjusted monitoring strategy fails to effectively improve the monitoring accuracy, then re-execute steps S401 to S405.

[0146] It should be noted that in step S406, the effect evaluation mainly compares the monitoring data before and after adjustment, and analyzes indicators such as the triggering frequency, false alarm rate, and missed alarm rate of the warning mechanism. If the evaluation result shows that the adjusted monitoring strategy fails to effectively improve the monitoring accuracy, the possible reasons include insufficient accuracy of the digital twin model, weak adaptability of the machine learning algorithm, or unreasonable selection of working condition parameters, etc. For these problems, data can be re-collected, model parameters can be optimized, the machine learning algorithm can be improved, or the selection range of working condition parameters can be adjusted to improve the accuracy and reliability of the monitoring strategy. Through continuous iteration and optimization, this embodiment can achieve precise monitoring and warning of the cable insulation performance, providing a strong guarantee for the safe operation of the cable.

[0147] Embodiment 2

[0148] This embodiment also provides an electronic device. Refer to Figure 2 , including a memory 404 and a processor 402. A computer program is stored in the memory 404, and the processor 402 is configured to run the computer program to execute the steps in any of the above embodiments of the on-line monitoring method for the insulation performance of power cables.

[0149] Specifically, the above-mentioned processor 402 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC for short), or may be configured as one or more integrated circuits implementing the embodiments of the present application.

[0150] Among them, the memory 404 may include a mass storage device 404 for data or instructions. By way of example and not limitation, the memory 404 may include a hard disk drive (HDD), a floppy disk drive, a solid state drive (SSD), a flash memory, an optical disc, a magneto-optical disc, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. In a suitable case, the memory 404 may include removable or non-removable (or fixed) media. In a suitable case, the memory 404 may be internal or external to the data processing device. In a particular embodiment, the memory 404 is a non-volatile memory. In a particular embodiment, the memory 404 includes a read-only memory (ROM) and a random access memory (RAM). In a suitable case, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically alterable ROM (EAROM), or a flash memory, or a combination of two or more of these. In a suitable case, the RAM may be a static random access memory (SRAM) or a dynamic random access memory (DRAM), where the DRAM may be a fast page mode dynamic random access memory (FPMDRAM), an extended data output dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0151] The memory 404 can be used to store or cache various data files required for processing and / or communication, as well as possible computer program instructions executed by the processor 402.

[0152] By reading and executing the computer program instructions stored in the memory 404, the processor 402 implements any one of the distribution methods for distributing data based on the rule engine encapsulation in the above embodiments.

[0153] Optionally, the above electronic device may further include a transmission device 406 and an input / output device 408. Among them, the transmission device 406 is connected to the above processor 402, and the input / output device 408 is connected to the above processor 402.

[0154] The transmission device 406 can be used to receive or send data via a network. Specific examples of the above network may include wired or wireless networks provided by the communication provider of the electronic device. In one example, the transmission device includes a network adapter (Network Interface Controller, abbreviated as NIC), which can be connected to other network devices through a base station and thus communicate with the Internet. In one example, the transmission device 406 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0155] The input / output device 408 is used to input or output information. In this embodiment, the input information can be various configuration information of the encapsulation component, etc., and the output information can be data to be distributed, etc.

[0156] Optionally, in this embodiment, the above processor 402 can be set to execute the following steps through a computer program:

[0157] Create an engineering service on the spring system, introduce the encapsulation component in the engineering service, and the encapsulation component depends on the dependency library as a dependency file. The encapsulation component at least includes a rule input configuration table for configuring rules, a condition configuration table for configuring rule conditions, and a sending configuration table for configuring the http protocol;

[0158] After starting the distribution service of the engineering service, add the encapsulation component to the IOC container;

[0159] The engineering service loads the rule configuration of the current data rule into the memory based on the encapsulation component, generates the current data rule based on the rule, adds the current data rule to the rule group, and registers the rule group to the bean of the IOC container;

[0160] Create an externally exposed access trigger interface on the engineering service and define the execution method of the rule engine, which, after being triggered, executes the data distribution task from the memory according to the rule configuration.

[0161] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation manners, and will not be elaborated herein.

[0162] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for online monitoring of insulation performance of power cables, characterized in that: The method comprises: S1. constructing a multimodal sensor network, and synchronously collecting multidimensional cable data based on the multimodal sensor network, wherein the multidimensional cable data includes partial discharge pulse data, distributed temperature field data, dielectric loss factor, and chemical characteristic gas data; S2. Establishing an electromagnetic-thermal-chemical multi-physics field coupling model, inputting the multi-dimensional cable data into the electromagnetic-thermal-chemical multi-physics field coupling model, and analyzing the multi-dimensional evolution law of the cable insulation state in real time; The electromagnetic field model is coupled with the heat conduction model, and the heat conduction model is coupled with the chemical reaction field model to obtain an electromagnetic-thermal-chemical multi-physics field coupling model; The expression of the electromagnetic-thermal-chemical multi-physics field coupling model is: ， ， ； in, represents the Hamiltonian operator, represents the magnetic permeability of the material, represents the magnetic vector potential, represents the conductivity, represents the partial derivative of the magnetic vector potential with respect to time, represents the source current density, represents the material density, represents the specific heat capacity, Indicates temperature, represents thermal conductivity, represents the Joule heat term, represents the concentration of chemical products, represents the diffusion coefficient, represents the frequency factor, represents the activation energy, is the gas constant, represents the reaction kinetic function; S3. Establish a hybrid deep learning model, use the hybrid deep learning model to extract the spatiotemporal features in the multi-dimensional evolution law, monitor the insulation defects of the cable based on the spatiotemporal features, and predict the remaining service life of the cable; S4. Based on the spatiotemporal characteristics, a digital twin method is used to dynamically generate a warning threshold, and the monitoring strategy is adaptively adjusted based on the warning threshold and operating condition parameters.

2. The method for online monitoring of insulation performance of power cables according to claim 1, characterized in that: In step S1, synchronously collecting multi-dimensional cable data based on the multimodal sensor network includes: S101, using a partial discharge detection system combining a differential Rogowski coil array and a high-frequency current transformer to collect partial discharge pulse data; S102, a distributed optical fiber temperature measurement system based on the Raman scattering principle collects distributed temperature field data; S103, collecting dielectric loss factor using a dielectric loss factor monitoring module including a programmable frequency source and an orthogonal sampling phase-sensitive demodulation circuit; S104, collecting chemical characteristic gas data using a gas monitoring unit composed of a nanotube array gas sensor and an optical fiber evanescent wave sensor.

3. The method for online monitoring of insulation performance of power cables according to claim 1, characterized in that: The establishment of the electromagnetic-thermal-chemical multi-physics field coupling model in step S2 includes: S201, establishing an electromagnetic field model using the partial discharge pulse data; S202, establishing a heat conduction model using the distributed temperature field data and dielectric loss factor; S203, using the chemical characteristic gas data to establish a chemical reaction field model; S204, coupling the electromagnetic field model with the heat conduction model, and coupling the heat conduction model with the chemical reaction field model to obtain an electromagnetic-thermal-chemical multi-physics field coupling model; S205, performing a confidence assessment on the electromagnetic-thermal-chemical multi-physics field coupling model, and optimizing the electromagnetic-thermal-chemical multi-physics field coupling model based on the assessment result.

4. The method for online monitoring of insulation performance of power cables according to claim 1, characterized in that: Establishing the hybrid deep learning model in step S3 includes: S301, performing spatiotemporal alignment and standardization processing on the multi-dimensional cable data to construct a tensor input including space-time dimensions; S302, constructing a spatial encoder based on a combination of a residual network and a dilated convolution, and using the spatial encoder to extract a spatial feature vector from the tensor input; S303, cascading a bidirectional long short-term memory network and a gated recurrent unit to obtain a time encoder, and using the time encoder to extract a time series feature vector of the insulation performance in the tensor input; S304, dynamically weighting and fusing the spatial feature vector and the temporal feature vector based on the cross-modal attention fusion module to obtain spatiotemporal features; S305, constructing a cable structure life prediction model using a graph convolutional network based on the spatiotemporal features; S306, constructing a cable insulation defect recognition model using a lightweight classifier based on the spatiotemporal features, and identifying the type of cable insulation defects based on the cable insulation defect recognition model; S307, respectively training and verifying the cable structure life prediction model and the cable insulation defect identification model; S308. Integrate the trained cable structure life prediction model and cable insulation defect recognition model to obtain a hybrid deep learning model.

5. The method for online monitoring of insulation performance of power cables according to claim 4, characterized in that: In step S302, the time encoder extracts the time series feature vector of the insulation performance in the tensor input as follows: ， ； in, represents the initial layer spatial feature map, Represents the spatial distribution data output by the electromagnetic-thermal-chemical multi-physics field coupling model, Indicates Layer spatial feature map, represents the activation function, Indicates The dilated convolution kernel of the layer, represents the dilated convolution operation, Indicates Layer spatial feature map, Indicates Bias of the layer; In step S303, the expression for extracting the time series feature vector of the insulation performance in the tensor input by using the time encoder is: ， ， ， ； in, Represents the time step The hidden state vector of the forward LSTM, represents the long short-term memory network, Represents the time step The input vector is the time series output of the electromagnetic-thermal-chemical multi-physics field coupling model. Represents the time step The hidden state vector of the forward LSTM, Represents the time step The hidden state vector of the LSTM before and after, Represents the time step The hidden state vector of the LSTM before and after, represents the bidirectional long short-term memory network at time step The concatenated hidden state of Represents a splicing operation, represents the gated recurrent unit at time step The hidden state of represents the gated recurrent unit operation, represents the gated recurrent unit at time step The hidden state of .

6. The method for online monitoring of insulation performance of power cables according to claim 4, characterized in that: In step S304, the expression of the cross-modal attention fusion module dynamically weighted fusion of spatial feature vectors and temporal feature vectors is: ， ， ， ； in, represents the spatial eigenvector after projection, represents the learned projection matrix, which is used to map the spatial feature vector and the temporal feature vector to a common dimension, Indicates global average pooling of the spatial feature map. represents the bias term of the projection layer, represents the time series feature vector after projection, The learned projection matrix is ​​used to map the temporal and spatial feature vectors to a common dimension, Indicates the time dimension pooling of the time series feature vector. represents the bias term of the projection layer, represents the dynamically generated attention weight vector, represents the activation function, represents the transpose of the projected spatial eigenvector, represents the learnable attention matrix, represents the scaling factor, Represents the spatiotemporal characteristics, Represents element-wise multiplication.

7. The method for online monitoring of insulation performance of power cables according to claim 4, characterized in that: In step S3, the expression for monitoring the insulation defects of the cable based on the spatiotemporal characteristics is: ； in, represents the predicted probability vector, represents the activation function, represents the classification layer parameters, Represents the spatiotemporal characteristics, Represents the bias vector of the classification layer; Monitoring the cable based on the spatiotemporal characteristics to predict the remaining service life of the cable in step S3 includes: ， ， ， ， ； in, represents the encoded hidden state sequence, Indicates encoding the input feature matrix, Represents the time step The normalized weight of represents the transpose of the feature transformation matrix, represents the feature transformation matrix, Represents the time step The hidden state sequence of represents the bias term of attention calculation, Represents the time step The hidden state sequence of represents the context vector, represents the predicted original life value, represents the decoding weight matrix, represents the decoding bias term, represents the life expectancy prediction value, represents the physical attenuation factor, Indicates the insulation aging rate coefficient, Represents the current time step.

8. The method for online monitoring of insulation performance of power cables according to claim 1, characterized in that: In step S4, the monitoring strategy is adaptively adjusted based on the warning threshold and the operating condition parameters, including: S401. Construct a digital twin model based on the multi-dimensional evolution law of cable insulation status; S402, the digital twin model dynamically generates a warning threshold of the cable insulation performance using a machine learning method based on the currently collected multi-dimensional cable data and the cable operating parameters; S403, when the insulation defect of the monitored cable approaches the warning threshold of the cable insulation performance, triggering the warning mechanism; S404, if the insulation defect of the monitoring cable is far away from the warning threshold, reduce the monitoring frequency; S405. If the insulation defect of the monitoring cable approaches or exceeds the warning threshold, increase the monitoring frequency; S406: Evaluate the effect of the adjusted monitoring strategy. If the evaluation result shows that the adjusted monitoring strategy fails to effectively improve the monitoring accuracy, re-execute steps S401 to S405.

9. An electronic device comprising a memory and a processor, characterized in that: The memory stores a computer program, and the processor is configured to run the computer program to execute the method for online monitoring of insulation performance of a power cable as claimed in any one of claims 1 to 8.

Images