Ultra-high voltage GIS graphene-copper-based composite diversion assembly fault diagnosis method and equipment

Abstract

The invention relates to the technical field of insulation switch detection, and provides an extra-high voltage GIS graphene-copper-based composite diversion assembly fault diagnosis method and equipment. Dynamic calibration and anti-interference data acquisition are performed on a target diversion component to obtain a multi-source calibration data set, time-space-frequency domain fusion is performed in combination with a Tucker decomposition mode to obtain a cross-modal coupling feature set, and time-frequency-space domain joint feature extraction and coding are performed on the cross-modal coupling feature set to obtain a joint vector. Fault classification is carried out in combination with an attribute graph attention network model to obtain fault category labels, multi-physics field coupling simulation is carried out according to the fault category labels and a multi-source calibration data set to obtain a fault area coordinate set, and finally residual life prediction is carried out through a dynamic Bayesian network model to obtain residual life probability distribution and a maintenance instruction. According to the invention, through multi-modal data fusion, depth feature extraction, intelligent fault classification, simulation auxiliary diagnosis, life prediction and intelligent operation and maintenance decision, the accuracy of fault detection is improved.

Description

Technical Field

[0001] The present application relates to the technical field of insulating switch detection, and in particular to a method and device for diagnosing faults of ultra-high voltage GIS graphene-copper-based composite guide components. Background Art

[0002] Ultra-high voltage gas-insulated switchgear (UHV) is a vital component of modern power systems and is widely used in ultra-high voltage and ultra-high voltage transmission projects. Its core component, the current guide assembly, is primarily responsible for current transmission and electric field distribution regulation, directly impacting the equipment's reliability and operational stability. In recent years, with the application of graphene-copper composite materials, these current guide assemblies have achieved significant improvements in conductivity, corrosion resistance, and thermal stability. However, under extreme operating conditions, they can still experience faults such as local overheating, stress concentration, electric field distortion, and insulating gas decomposition. In severe cases, these faults can even cause equipment damage and grid failures. Therefore, conducting efficient and accurate fault diagnosis of UHV GIS graphene-copper composite current guide assemblies is of great significance for ensuring grid security, improving equipment lifespan, and optimizing operation and maintenance strategies.

[0003] Currently, fault diagnosis methods for UHV GIS flow guide components primarily include thermal imaging analysis, partial discharge detection, and acoustic emission monitoring, typically performed using a single physical field. Existing methods typically rely on empirical judgment or analysis of single physical quantity anomalies to locate faults, making it difficult to achieve high-precision spatial localization. For example, thermal imaging can only provide surface temperature distribution, while finite element simulation, lacking high-precision boundary conditions, can produce results that deviate significantly from actual conditions. Summary of the Invention

[0004] In view of this, the present application provides a method and device for diagnosing faults of ultra-high voltage GIS graphene-copper-based composite guide components to solve the problem of large deviation in fault diagnosis of guide components.

[0005] In a first aspect, the present application provides a method for diagnosing a fault of a UHV GIS graphene-copper-based composite flow guide assembly, the method comprising: Perform dynamic calibration and anti-interference data acquisition and processing on the target diversion component to obtain a multi-source calibration data set; Performing a time-space-frequency domain fusion process on the multi-source calibration dataset by a preset Tucker decomposition method to obtain a cross-modal coupling feature set; Performing time-frequency-spatial domain joint feature extraction and encoding processing on the cross-modal coupling feature set to obtain a joint vector; Performing fault classification processing on the joint vector through a preset attribute graph attention network model to obtain a fault category label; Performing multi-physics field coupling simulation processing according to the fault category label and the multi-source calibration data set to obtain a fault area coordinate set; The remaining life prediction process is performed based on the fault area coordinate set and preset historical degradation data through a preset dynamic Bayesian network model to obtain the remaining life probability distribution and maintenance instructions.

[0006] In an optional embodiment, the dynamically calibrating the target diversion component and performing anti-interference data acquisition processing to obtain a multi-source calibration data set includes: Performing infrared thermal imaging on the target guide component to obtain surface grayscale values, and performing nonlinear temperature calibration on the surface grayscale values ​​to obtain calibrated temperature field data; Acquiring an acoustic emission signal from the target flow guide component to obtain an original acoustic emission signal, and performing frequency domain inverse filtering compensation processing on the original acoustic emission signal to obtain compensated optimized acoustic emission data; Performing microstrain acquisition on the target flow guide component to obtain measured strain data, and performing thermal expansion correction processing on the measured strain data according to the calibration temperature field data to obtain stress field data; Performing a partial discharge test on the target flow guide component to obtain an original discharge pulse signal, and performing time domain synchronous compensation processing on the original discharge pulse signal to obtain time domain optimized discharge pulse phase data; Performing insulation gas decomposition mass spectrum peak data collection and linear calibration processing on the target flow guide component to obtain target gas concentration data; Data fusion processing is performed on the calibration temperature field data, the compensated optimized acoustic emission data, the stress field data, the time domain optimized discharge pulse phase data and the target gas concentration data to obtain the multi-source calibration data set.

[0007] In an optional embodiment, performing space-time-frequency fusion processing on the multi-source calibration dataset by a preset Tucker decomposition method to obtain a cross-modal coupling feature set includes: Performing unified spatiotemporal mapping processing on the calibrated temperature field data, the compensated optimized acoustic emission data, the stress field data, and the time-domain optimized discharge pulse phase data to obtain a four-dimensional heterogeneous tensor; Factor decomposition is performed on the four-dimensional heterogeneous tensor using a preset Tucker decomposition method to obtain an original core tensor and a factor matrix set; Performing energy truncation processing on the original core tensor according to a preset energy proportion threshold to obtain a compressed feature tensor; A cross-modal coupling feature fusion storage process is performed on the compressed feature tensor and the factor matrix set to obtain the cross-modal coupling feature set.

[0008] In an optional embodiment, performing time-frequency-spatial domain joint feature extraction and encoding processing on the cross-modal coupling feature set to obtain a joint vector includes: Performing spatial mode reconstruction processing on the cross-modal coupling feature set to obtain gradient field data of each modality; Performing gradient modulus calculation processing on each modal gradient field data to obtain comprehensive gradient modulus data; Performing wavelet packet decomposition processing on the compensated optimized acoustic emission data to obtain multi-layer wavelet packet coefficient data; Performing energy density calculation processing on the multi-layer wavelet packet coefficient data to obtain time-frequency energy density data; Vectorized splicing processing is performed on the comprehensive gradient modulus data, the time-frequency energy density data, and the target gas concentration data to obtain the joint vector.

[0009] In an optional embodiment, performing fault classification processing on the joint vector using a preset attribute graph attention network model to obtain a fault category label includes: Performing node division processing on the joint vector to obtain an attribute vector of each monitoring point; Performing neighborhood relationship construction processing on the attribute vector according to preset heat conduction and mechanical connection rules to obtain edge weights between nodes; Calculate the attention coefficient according to the edge weights between the nodes to obtain the attention coefficient; The attribute graph attention network model performs weighted aggregation processing on the attribute vectors of the neighborhood nodes according to the attention coefficient to obtain the node feature vector; Softmax classification is performed on the node feature vector to obtain the fault category label.

[0010] In an optional embodiment, performing multi-physics field coupling simulation processing according to the fault category label and the multi-source calibration data set to obtain a fault area coordinate set includes: Determining a corresponding fault physical field type according to the fault category label, and performing target physical field data screening processing on the multi-source calibration data set to obtain a target physical field data subset; Performing three-dimensional space grid division processing on the target physical field data subset to obtain a multi-physics field coupling simulation calculation grid; Performing thermal-electrical-mechanical-gas multi-field joint boundary condition assignment processing on the multi-physics field coupling simulation calculation grid to obtain a boundary constraint condition matrix; Performing finite element solution processing on the multi-physics field coupling simulation calculation grid according to the boundary constraint matrix to obtain fault response distribution field data; Performing isosurface extraction processing on the fault response distribution field data to obtain distribution information of high fault response areas; A spatial coordinate analysis process is performed based on the distribution information of the high fault response area to obtain the fault area coordinate set.

[0011] In an optional embodiment, the performing of remaining life prediction processing based on the fault area coordinate set and preset historical degradation data using a preset dynamic Bayesian network model to obtain a remaining life probability distribution and maintenance instructions includes: Performing spatial matching processing on the historical degradation data according to the fault area coordinate set to obtain target fault evolution data; Performing time-series segmented modeling processing on the target fault evolution data to obtain a multi-stage degradation state sequence; Performing state transition probability calculation processing according to the multi-stage degradation state sequence to obtain a time series state transition matrix; Performing latent variable joint distribution fitting processing on the time series state transfer matrix to obtain a dynamic Bayesian network prior parameter set; Performing latent variable joint distribution fitting processing on the time series state transfer matrix to obtain a dynamic Bayesian network prior parameter set; Performing decision threshold analysis on the remaining life probability distribution to obtain a maintenance risk assessment indicator; A maintenance instruction generation process is performed according to the maintenance risk assessment indicator to obtain the maintenance instruction.

[0012] A second aspect of the present application provides a fault diagnosis device for a UHV GIS graphene-copper-based composite flow guide assembly, the device comprising: The data acquisition module is used to dynamically calibrate the target diversion component and perform anti-interference data acquisition and processing to obtain a multi-source calibration data set; A data fusion module is used to perform time-space-frequency fusion processing on the multi-source calibration data set through a preset Tucker decomposition method to obtain a cross-modal coupling feature set; a joint extraction module, configured to perform time-frequency-spatial domain joint feature extraction and encoding processing on the cross-modal coupling feature set to obtain a joint vector; A fault classification module is used to perform fault classification processing on the joint vector through a preset attribute graph attention network model to obtain a fault category label; a fault simulation module, configured to perform multi-physics field coupling simulation processing according to the fault category label and the multi-source calibration data set to obtain a fault area coordinate set; The life prediction module is used to perform remaining life prediction processing based on the fault area coordinate set and preset historical degradation data through a preset dynamic Bayesian network model to obtain remaining life probability distribution and maintenance instructions.

[0013] The third aspect of the present application provides an electronic device, which includes a memory, a processor, and a computer program stored on the memory and runnable on the processor. When the processor executes the computer program, it implements the steps of the ultra-high voltage GIS graphene-copper-based composite guide assembly fault diagnosis method as described above.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above-mentioned method for diagnosing faults of ultra-high voltage GIS graphene-copper-based composite guide components are implemented.

[0015] In summary, this application has at least the following beneficial technical effects: 1. Through data optimization strategies such as nonlinear temperature calibration, inverse filtering compensation, thermal expansion correction, and time domain synchronization compensation, the impact of environmental noise and signal distortion on data quality is reduced, making the final acquired data more reliable and ensuring the stability and accuracy of subsequent analysis.

[0016] 2. Use thermal-electrical-mechanical-gas multi-field coupling simulation to accurately calculate the spatial coordinates of the fault area through finite element solution and boundary condition assignment.

[0017] 3. The topological information of the thermal-electrical-mechanical interaction inside the diversion component is used to improve the accuracy of fault identification. The feature weights can be dynamically adjusted to avoid the fixed weighting problem, making the classification more adaptive.

[0018] 4. Using a dynamic Bayesian network model, fault zone coordinates and historical degradation data are used to construct a multi-stage degradation model and calculate state transition probabilities to predict the remaining life of GIS diversion components. This allows for more accurate life assessment by adjusting predictions based on real-time monitoring data. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 This is a flow chart of a fault diagnosis method for a UHV GIS graphene-copper-based composite guide assembly provided in an embodiment of the present application; Figure 2 This is a functional module diagram of a fault diagnosis device for a UHV GIS graphene-copper-based composite flow guide assembly provided in an embodiment of the present application; Figure 3 It is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0022] like Figure 1 FIG2 is a flow chart of a method for diagnosing a fault of a UHV GIS graphene-copper composite flow guide assembly provided in an embodiment of the present application. The method for diagnosing a fault of a UHV GIS graphene-copper composite flow guide assembly provided in an embodiment of the present application comprises the following steps.

[0023] Step S1: Dynamically calibrate the target diversion component and perform anti-interference data acquisition and processing to obtain a multi-source calibration data set.

[0024] The system collects raw data from the target guide assembly surface and pre-processes the data to ensure the accuracy of the data used for fault diagnosis. The multi-modal sensor array includes infrared thermal imaging sensors, acoustic emission sensors, micro-strain gauges, ultra-high frequency (UHF) sensors, and a gas mass spectrometer.

[0025] First, a blackbody radiation source (set to a temperature range of 20°C-300°C) is placed on the surface of the target deflector assembly, allowing the infrared thermal imaging sensor to capture the infrared thermal imaging grayscale values ​​of the target deflector assembly at different temperatures. Because there is a nonlinear relationship between grayscale values ​​and actual temperature, the collected grayscale values ​​need to be calibrated nonlinearly to convert the collected grayscale data into accurate temperature values ​​to eliminate the nonlinear error of the sensor itself. The infrared thermal imaging temperature calibration process uses the polynomial fitting method shown below: Where T(x,y) is the temperature value at the pixel coordinate (x,y). G(x,y) is the grayscale value collected at the pixel coordinate (x,y), and the value range is usually 0 to 255. 0、 a 1、 a2 and a3 are the constant term, first-order, second-order and third-order coefficients, respectively, which are obtained by fitting the standard temperature and grayscale data collected from the blackbody radiation source using the least squares method.

[0026] Secondly, the original signal collected by the acoustic emission sensor will be affected by the imbalance of the sensor's own frequency response, so frequency domain inverse filtering compensation is required. The compensation process first measures the sensor's frequency response function. , and design the inverse filter function according to the frequency response function, so as to perform frequency domain inverse filtering compensation according to the inverse filter function. The frequency domain inverse filtering compensation processing process can be expressed as shown in the following formula: Among them, S cal (t) To compensate and optimize acoustic emission data, the effective bandwidth is increased to 0.1-3MHz. =10 -6 is a minimum constant to avoid the denominator in the inverse filter function being zero. By constructing the inverse filter function, the true frequency band characteristics of the original signal can be restored after the acoustic emission signal is filtered in the frequency domain.

[0027] At the same time, the micro-strain gauge used to measure the strain data of the target guide component under stress will produce a thermal expansion effect due to the temperature of the environment in which it is located, resulting in false strain caused by temperature changes mixed in the measurement results. Therefore, it is necessary to perform thermal expansion decoupling processing on the measured strain based on the temperature field data. The micro-strain gauge thermal expansion decoupling processing can be expressed as follows: in, is the true strain value, which is the strain after temperature compensation. is the strain data directly measured by the micro strain gauge. α is the thermal expansion coefficient of the material (e.g., 2.3x10 -6 / ℃), used to express the strain change caused by the material when the temperature changes by one unit. cal is the actual temperature field data obtained from the infrared thermal imager after nonlinear calibration. T0 is the reference temperature (e.g., 25°C). Thermal expansion decoupling of the microstrain gauge removes the additional strain caused by temperature changes, thereby obtaining true mechanical strain data. This data can then be used to calculate stress field data, ensuring the accuracy of fault location analysis.

[0028] The discharge pulse signals collected by the UHF sensor have phase deviations during multi-channel transmission. To ensure that all signals have a unified time base, a clock synchronization method is used to perform phase compensation on the signals. The time domain synchronization compensation process can be expressed as follows: in, is the phase of the discharge pulse signal after synchronization. is the original discharge pulse signal phase. is the signal frequency, typically in the range of 0.3 to 3 GHz. Δt is the actual measured propagation delay between channels, typically less than 1 nanosecond. 2π is the constant used to convert the delay into a phase compensation value. Time synchronization is performed on the signals collected from each channel to ensure phase consistency, thereby providing accurate alignment in the time domain for subsequent joint analysis of signal characteristics.

[0029] It should be understood that the insulating gas sulfur hexafluoride used in gas-insulated switchgear will decompose into sulfur dioxide and hydrogen fluoride due to hydrolysis reactions caused by discharge faults and high temperature conditions. The gas mass spectrometer collects mass spectrum peak data of sulfur hexafluoride decomposition products (for example, I m / z=64 The corresponding sulfur dioxide peak and I m / z=20 The corresponding hydrogen fluoride peak is converted into gas concentration data using a preset linear calibration formula. The quantitative calibration process of the gas mass spectrometer can be expressed as follows: Among them, C SO2 、C HF Represent the gas concentrations of sulfur dioxide and hydrogen fluoride respectively. Represents the mass spectrum peak intensity at mass-to-charge ratios of 64 and 20, respectively. SO2 , K HF The peak intensity data measured by the mass spectrometer is converted into actual gas concentration, thereby reflecting the content of sulfur hexafluoride decomposition products in the insulating material of the guide component.

[0030] After rigorous preprocessing, the resulting multi-source calibration data set includes calibrated temperature field data, compensated and optimized acoustic emission data, stress field data, time-domain optimized discharge pulse phase data, and target gas concentration data. This provides accurate, clean, and physically meaningful raw input information for subsequent multimodal data fusion and fault diagnosis. Each processing step addresses inherent measurement errors, nonlinearities, and interference in the corresponding sensor data, correcting them using mathematical models and engineering methods to ensure the authenticity and reliability of the final data.

[0031] Step S2: performing space-time-frequency fusion processing on the multi-source calibration dataset using a preset Tucker decomposition method to obtain a cross-modal coupling feature set.

[0032] Furthermore, the multi-source calibration datasets were unified to address differences in spatial resolution, sampling rate, and data scale among the various sensor data. Specifically, by constructing a four-dimensional heterogeneous tensor, the calibration temperature field data, compensated and optimized acoustic emission data, stress field data, and time-domain optimized discharge pulse phase data were mapped onto a unified spatiotemporal and frequency-domain grid, providing a structured data foundation for subsequent feature extraction.

[0033] First, the obtained calibration temperature field data, compensation optimization acoustic emission data, stress field data and time domain optimization discharge pulse phase data are uniformly mapped in time and space. This process is based on the preset spatial grid division (for example, 100×100 two-dimensional grid), the number of frequency sub-bands of acoustic emission data (for example, 256 sub-bands) and the time series sampling points (for example, 10 4 The process of constructing a four-dimensional heterogeneous tensor can be expressed as follows: in, is the spatial coordinate (x, y), frequency subband And the tensor elements at time t. T cal (x, y, t) is the calibration temperature field data. S cal (f, t) is the compensated optimized acoustic emission data. σ(x, y, t) is the stress field data. Optimize the discharge pulse phase data for the time domain. x, y, f, and t are discrete indices of space, frequency, and time, respectively. Mapping all modal data into a single tensor resolves inconsistencies in data scale and sampling rate, providing a unified input for the subsequent Tucker decomposition.

[0034] Secondly, the constructed four-dimensional heterogeneous tensor is subjected to Tucker decomposition to compress the high-dimensional data into a low-dimensional core tensor and extract the coupling characteristics between the modes. The Tucker decomposition process can be expressed as follows: in, is the constructed four-dimensional heterogeneous tensor. is the original core tensor, and its size is usually set to the feature tensor after dimensionality reduction (for example, 20×20×50×100). (1) 、U (2) 、U (3) 、U (4) Corresponding to the factor matrices in the horizontal space, vertical space, frequency domain and time direction respectively, these matrices satisfy the orthogonal constraints. n represents the product of a tensor and a matrix at the nth mode. ε is the residual tensor, representing the noise not captured during the decomposition process. Tucker decomposition is used to decompose high-dimensional data into low-dimensional core tensors and factor matrices, thereby extracting the main coupling features across modalities while effectively compressing the data volume.

[0035] After obtaining the original core tensor, energy truncation is performed to remove redundant and low-energy components. This truncation is based on the energy proportions of the components within the core tensor. Typically, a cumulative energy threshold of 95% is set, and components below this threshold are discarded, retaining the primary features. By removing redundant, low-energy components from the core tensor and retaining the primary features, the core tensor is compressed while maintaining the integrity of the cross-modal coupling features.

[0036] The resulting cross-modal coupling feature set consists of a compressed feature tensor and a set of factor matrices in each direction. This feature set encompasses the key characteristics of each sensor data in the temporal, spatial, and frequency domains, while eliminating interference caused by inconsistencies in sensor data, laying the foundation for subsequent joint feature extraction and encoding.

[0037] Step S3: performing time-frequency-spatial domain joint feature extraction and encoding processing on the cross-modal coupling feature set to obtain a joint vector.

[0038] Furthermore, through the joint feature extraction method of time-frequency and spatial domains, a joint vector is constructed that can comprehensively reflect the physical characteristics of the diversion component failure, which comprehensively reflects the temperature, stress, discharge and gas decomposition state of the diversion component, and provides a comprehensive feature representation that can capture the coupling characteristics of multiple physical fields.

[0039] First, the obtained transverse spatial factor matrix and compressed feature tensor are subjected to spatial pattern reconstruction to extract the gradient field information of each mode in the spatial dimension. The spatial pattern reconstruction process can be expressed as follows: in, is the first point at the spatial coordinate (x, y) Modal gradient field data. is the element in the xth row and ith column of the first factor matrix (i.e., the transverse spatial factor matrix), which is used to reflect the spatial modal distribution. The factor matrix is ​​used to compress the components corresponding to the i-th spatial mode and the k-th mode in the feature tensor, with the last two dimensions corresponding to the frequency domain and time. By combining the factor matrix with the information in the core tensor, the gradient field pattern at each spatial location is reconstructed, providing a spatial reference for detecting local temperature or stress mutations.

[0040] Secondly, the gradient modulus length of each modal gradient field data is calculated to generate a single quantity that describes the degree of gradient change in the physical space. The gradient modulus length calculation can be expressed as the following formula: in, It is the comprehensive gradient modulus at the spatial coordinate (x, y), reflecting the strength of the temperature or stress gradient. The gradient field data of each mode is integrated into a scalar value, which is convenient for describing the sudden change characteristics of local physical quantities and providing an intuitive spatial index for fault location.

[0041] At the same time, wavelet packet decomposition technology is used to extract time-frequency features from the compensated optimized acoustic emission data. Wavelet packet decomposition can subdivide the signal at different frequency and time scales to obtain multi-level frequency domain coefficients. The wavelet packet decomposition process can be expressed as follows: W m,n (t,k) Among them, W m,n (t, k) represents the decomposition coefficients for the mth layer and the nth node at time t in the wavelet packet decomposition, where k may represent a specific frequency index. m represents the number of decomposition layers (for example, set to 6 layers). n represents the number of wavelet packet nodes (for example, 64 nodes). t represents the time series index. Wavelet packet decomposition provides a detailed energy distribution of the acoustic emission signal across different frequency bands, providing the time-frequency data foundation for subsequent energy density calculations.

[0042] Subsequently, the energy density calculation process is performed on each node coefficient obtained by wavelet packet decomposition. The energy density calculation process can be expressed as follows: Among them, E m,n (t) is the energy density at the mth layer and nth node at time t. m,n (t, k) is the wavelet packet decomposition coefficient. By calculating the energy distribution of each node in the time domain, the energy concentration of the signal in each frequency band is reflected, which facilitates the identification of time-frequency information such as discharge characteristics.

[0043] Finally, the integrated gradient modulus data, time-frequency energy density data, and target gas concentration data are vectorized and spliced ​​together. This process splices the feature data from different sources into a one-dimensional joint vector in a predetermined order, forming the overall feature input. The vectorized splicing process can be expressed as follows: Where F is the final joint vector, representing the combination of all feature information. vec() is a vectorization operation that converts two-dimensional or multi-dimensional data into a one-dimensional vector. is the physical space gradient data. E m,n (t) is the time-frequency energy density data. C gas (t) is the target gas concentration data.

[0044] Step S4: perform fault classification processing on the joint vector through a preset attribute graph attention network model to obtain a fault category label.

[0045] It should be understood that the attribute graph attention network model can capture the physical connections between monitoring points in the target flow control component by constructing a monitoring point attribute graph, calculating similarity weights between nodes, aggregating node information using graph convolution, and finally obtaining fault category labels through softmax classification. For example, the conduction and mechanical connection characteristics of temperature gradients can be used to more accurately distinguish various fault states.

[0046] First, the joint vector is partitioned into nodes, which are divided into attribute vectors for each monitoring point. The attribute vector corresponding to each monitoring point records the information of the point in multiple dimensions such as temperature gradient, time-frequency energy, and gas concentration. The node partitioning process can be expressed as follows: Where F is the joint vector. i is the attribute vector of the i-th monitoring point. N is the total number of monitoring points. By dividing the overall joint vector into multiple local vectors, each vector corresponds to the multimodal features of a monitoring point, it is convenient to construct the subsequent graph structure.

[0047] Furthermore, based on the preset heat conduction and mechanical connection rules, the similarity between the attribute vectors of each monitoring point is measured to build a neighborhood relationship, thereby calculating the edge weights between nodes. The neighborhood relationship construction process can be expressed as follows: Among them, W ij is the edge weight between monitoring points i and j. Represent the temperature gradient data at monitoring points i and j respectively. Tis the temperature gradient similarity threshold, which is set to a fixed value (e.g., 5°C / mm). is an indicator function that takes the value 1 if monitoring points i and j are mechanically connected, and 0 otherwise. exp represents an exponential function. By calculating the similarity and physical connection between monitoring points, an edge weight matrix is ​​generated that reflects the neighborhood relationships of the monitoring points, providing the basis for graph convolution operations.

[0048] Subsequently, the attention coefficient needs to be calculated for the obtained edge weights between nodes. Through the preset attention mechanism, the concatenation information of each node's attribute vector is used to calculate the attention distribution of each node to the information of other nodes in its neighborhood. The attention coefficient calculation process can be expressed as follows: Among them, α ij For nodes For neighboring nodes The attention coefficient. is the concatenation of the attribute vectors of nodes i and j. a is the preset trainable attention vector. LeakyReLU(·) is the LeakyReLU activation function used for nonlinear transformation. N(i) is the node By calculating the information transmission weight between nodes, the graph network can dynamically adjust the information aggregation weight according to the similarity of each node's attributes.

[0049] Next, we use the attribute graph attention network model to perform graph convolution on the attribute vectors of each monitoring point. By weightedly aggregating the attribute vectors of nodes in each node's neighborhood according to the attention coefficient and processing them through the activation function, we obtain the updated node feature vector. The graph convolution weighted aggregation process can be expressed as follows: in, For nodes Updated feature vector. σ() is the activation function, usually ReLU or LeakyReLU. α ij is the attention coefficient of node i to neighboring node j. W is the graph convolution transformation matrix, which is used to linearly transform node features. By weightedly aggregating the information of neighboring nodes, each node's features are updated to form a more discriminative feature representation, providing data support for fault status classification.

[0050] Finally, the updated node feature vector is subjected to Softmax classification processing, the node features are mapped to the predetermined fault category space, and the final fault category label is obtained by calculating the probability distribution of each category. The Softmax classification process can be expressed as follows: in, is the probability that node i belongs to category k. k is the weight matrix for category k. K is the total number of fault categories. The updated node feature vectors are mapped to a probability distribution using the Softmax function. This ultimately determines the fault category label for each monitoring point, providing a classification basis for subsequent fault area location and remaining life prediction.

[0051] Step S5: performing multi-physics field coupling simulation processing according to the fault category label and the multi-source calibration data set to obtain a fault area coordinate set.

[0052] Based on the fault classification results and a multi-source calibration dataset, a multi-physics coupled simulation was performed to identify the specific fault location within the flow guide assembly. By constructing a coupled electromagnetic, thermal, and mechanical model, the residual between the theoretical and measured temperature fields was calculated, and the coordinates of the fault location were extracted.

[0053] First, based on the multi-source calibration dataset and fault classification results, physical field data relevant to the fault state is selected. The multi-source calibration dataset is then subjected to target data screening to obtain a target physical field data subset for simulation. This data subset includes temperature fields, stress fields, and other physical quantities closely related to the fault. Next, the target physical field data subset is meshed in three dimensions to construct a computational grid for multi-physics coupled simulation, ensuring that the spatial resolution can capture subtle changes in the localized fault area.

[0054] Subsequently, boundary conditions are set for the multi-physics coupled simulation, and the computational grid is solved using an electromagnetic-thermal-mechanical coupled model. Specifically, for heat conduction, the internal heat conduction model of the flow guide component is used to calculate the temperature distribution using the finite element method; for electromagnetics, the distribution of the electric and magnetic fields is calculated using the Maxwell equations; and for mechanics, the stress field changes are calculated using the stress-strain relationship. The multi-physics coupled simulation solution process can be expressed as follows: Where T is the temperature field; K is the heat transfer coefficient; is the gradient operator, which means partial derivative of spatial variables; is the resistivity; is the modulus of current density; β is the thermoelastic coupling coefficient, which describes the heat generation caused by stress changes; is the rate of change of stress over time. Multi-physics coupled simulation is used to describe the coupling relationship between electromagnetic, thermal, and mechanical fields. Solving the boundary conditions yields a theoretical temperature field distribution, providing theoretical predictions for fault location.

[0055] After the simulation is completed, the difference between the actual measured calibration temperature field data and the theoretical simulation temperature field is compared to calculate the residual field. The residual calculation process can be expressed as follows: Where ΔT(x,y) is the temperature residual at the spatial coordinate (x,y). sim (x,y) is the theoretical temperature field obtained by simulation calculation. T cal (x, y, t) represents the actual calibration temperature field data. Residual calculation is used to quantitatively describe the difference between the theoretical temperature field and the actual measured temperature field. Areas with large residuals usually correspond to fault areas.

[0056] After obtaining the residual field, we perform isosurface extraction on the residual data using statistical analysis methods. Using a statistical threshold method, we set the third quartile (Q3) and interquartile range (IQR) of the residual as the benchmark. We determine whether the residual exceeds the standard of Q3 + 1.5.IQR. Areas exceeding this standard are considered faulty. We then use statistical methods to identify abnormal residual areas and mark them as fault coordinates, providing fault location information for subsequent remaining life prediction.

[0057] Step S6: Using a preset dynamic Bayesian network model, perform remaining life prediction processing based on the fault area coordinate set and preset historical degradation data to obtain a remaining life probability distribution and maintenance instructions.

[0058] Based on the fault area location results and historical degradation data, the degradation process of the diversion component is modeled and predicted through a dynamic Bayesian network model combined with a particle filtering method. The final output is the remaining life probability distribution and maintenance instructions. The purpose is to quantitatively describe the transition process of the diversion component from a healthy state to a faulty state, and to predict future failure times through time series analysis.

[0059] First, spatially match the historical degradation data based on the fault region coordinate set. The historical degradation data from each monitoring point is matched to the fault region to obtain the target fault evolution data. The matched data includes the historical evolution curves of temperature residuals, stress changes, and gas concentrations. This matching process clearly defines the past evolution trajectory of the fault region, providing basic input data for subsequent modeling.

[0060] Next, the acquired target fault evolution data is subjected to time-series segmented modeling. This process divides the entire time series data into several stages, each corresponding to a different state of the diversion component degradation process, such as "healthy," "early degradation," and "late degradation." State changes within each stage can be described using a statistical model, resulting in a multi-stage degradation state sequence. By defining the state space during the degradation process, this provides the foundation for establishing hidden state transitions in the dynamic Bayesian network model.

[0061] After obtaining a multi-stage degradation state sequence, the transition probabilities between states are calculated based on historical data to construct a state transition matrix. The state transition matrix describes the probability of a component transitioning from one degradation state to another and serves as the basis for the state transition probabilities in the dynamic Bayesian network model. The state transition probability calculation process can be expressed as follows: in, The state at the previous moment is S t-1 When the current state is S t The conditional probability of state transitions in statistical historical data represents the transition probability calculated through statistical analysis of historical degradation data. By determining the transition probabilities between states, a state transition model in a dynamic Bayesian network is constructed, providing a probabilistic basis for state prediction.

[0062] Subsequently, the constructed state transition matrix is ​​fitted with the joint distribution of latent variables to determine the prior parameter set of the dynamic Bayesian network. This process uses the particle filter method to infer the state transition process in real time, generating a set of particles representing different degradation trajectories and updating the weight of each particle based on the observed data. The particle filter weight update process can be expressed as follows: in, is the weight of the i-th particle at time t. In state Observe the data below O t The likelihood probability of . To transfer from the state of particle j to particle The conditional probability of the state. N is the total number of particles. By updating the particle weights, each particle can adjust its probability based on the current observation data, ultimately approximating the probability distribution of the actual degradation process.

[0063] After obtaining the updated particle set and its weight distribution, we perform time series inference on each particle's state to calculate the probability distribution of its remaining lifetime. By statistically analyzing the distribution of the times at which each particle reaches the "late degradation" state in the future, we can determine the probability prediction of its remaining lifetime.

[0064] in, is the probability of failure (entering the late degradation state) within time t. is the weight of the i-th particle at time t. is an indicator function that takes the value 1 when particle i is in an advanced degradation state and 0 otherwise. The probability distribution of the remaining life is calculated based on the particle filtering results, thus providing a quantitative basis for maintenance decisions.

[0065] Finally, maintenance risk is assessed based on the remaining life probability distribution, and corresponding maintenance instructions are generated based on pre-set decision thresholds. These instructions include specific inspection times, maintenance plans, and preventative measures to ensure timely repair or replacement of the guide components before they reach critical degradation, thereby reducing the risk of accidents caused by failures.

[0066] This application is applied to the field of insulating switch detection technology. It obtains a multi-source calibration data set by dynamically calibrating the target diversion component and collecting anti-interference data. It combines the Tucker decomposition method to perform time-space-frequency domain fusion to obtain a cross-modal coupling feature set. It performs time-frequency-space domain joint feature extraction and encoding on the cross-modal coupling feature set to obtain a joint vector. It combines the attribute graph attention network model to perform fault classification to obtain the fault category label. It performs multi-physics field coupling simulation based on the fault category label and the multi-source calibration data set to obtain the fault area coordinate set. Finally, it uses the dynamic Bayesian network model to predict the remaining life to obtain the remaining life probability distribution and maintenance instructions. This application realizes high-precision, high-robustness, and high-intelligence GIS diversion component fault diagnosis through multi-modal data fusion, deep feature extraction, intelligent fault classification, simulation-assisted diagnosis, life prediction, and intelligent operation and maintenance decision-making. It not only improves the accuracy and reliability of fault detection, but also provides more scientific and reasonable maintenance decision support for operation and maintenance personnel, greatly improving the safety and service life of ultra-high voltage GIS equipment.

[0067] like Figure 2 , which is a functional module diagram of a fault diagnosis device for a UHV GIS graphene-copper-based composite guide assembly provided in an embodiment of the present application.

[0068] In some embodiments, the UHV GIS graphene-copper composite flow guide assembly fault diagnosis device 2 may include multiple functional modules composed of computer program segments. The computer program of each program segment in the UHV GIS graphene-copper composite flow guide assembly fault diagnosis device 2 may be stored in the memory of the server and executed by at least one processor to perform (see Figure 1 (Describe) the function of the fault diagnosis method of UHV GIS graphene-copper-based composite guide assembly.

[0069] In this embodiment, the UHV GIS graphene-copper composite flow guide assembly fault diagnosis device 2 can be divided into multiple functional modules based on the functions they perform. These functional modules may include: a data acquisition module 21, a data fusion module 22, a joint extraction module 23, a fault classification module 24, a fault simulation module 25, and a life prediction module 26. As used herein, a module refers to a series of computer program segments that can be executed by at least one processor and perform a fixed function, and are stored in a memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0070] The data acquisition module 21 is used to perform dynamic calibration and anti-interference data acquisition processing on the target diversion component to obtain a multi-source calibration data set.

[0071] In an optional embodiment, the data acquisition module 21 is specifically configured to: Performing infrared thermal imaging on the target guide component to obtain surface grayscale values, and performing nonlinear temperature calibration on the surface grayscale values ​​to obtain calibrated temperature field data; Acquiring an acoustic emission signal from the target flow guide component to obtain an original acoustic emission signal, and performing frequency domain inverse filtering compensation processing on the original acoustic emission signal to obtain compensated optimized acoustic emission data; Performing microstrain acquisition on the target flow guide component to obtain measured strain data, and performing thermal expansion correction processing on the measured strain data according to the calibration temperature field data to obtain stress field data; Performing a partial discharge test on the target flow guide component to obtain an original discharge pulse signal, and performing time domain synchronous compensation processing on the original discharge pulse signal to obtain time domain optimized discharge pulse phase data; Performing insulation gas decomposition mass spectrum peak data collection and linear calibration processing on the target flow guide component to obtain target gas concentration data; Data fusion processing is performed on the calibration temperature field data, the compensated optimized acoustic emission data, the stress field data, the time domain optimized discharge pulse phase data and the target gas concentration data to obtain the multi-source calibration data set.

[0072] The data fusion module 22 is configured to perform a time-space-frequency fusion process on the multi-source calibration dataset using a preset Tucker decomposition method to obtain a cross-modal coupling feature set.

[0073] In an optional embodiment, the data fusion module 22 is specifically configured to: Performing unified spatiotemporal mapping processing on the calibrated temperature field data, the compensated optimized acoustic emission data, the stress field data, and the time-domain optimized discharge pulse phase data to obtain a four-dimensional heterogeneous tensor; Factor decomposition is performed on the four-dimensional heterogeneous tensor using a preset Tucker decomposition method to obtain an original core tensor and a factor matrix set; Performing energy truncation processing on the original core tensor according to a preset energy proportion threshold to obtain a compressed feature tensor; A cross-modal coupling feature fusion storage process is performed on the compressed feature tensor and the factor matrix set to obtain the cross-modal coupling feature set.

[0074] The joint extraction module 23 is configured to perform time-frequency-spatial domain joint feature extraction and encoding processing on the cross-modal coupling feature set to obtain a joint vector.

[0075] In an optional embodiment, the joint extraction module 23 is specifically configured to: Performing spatial mode reconstruction processing on the cross-modal coupling feature set to obtain gradient field data of each modality; Performing gradient modulus calculation processing on each modal gradient field data to obtain comprehensive gradient modulus data; Performing wavelet packet decomposition processing on the compensated optimized acoustic emission data to obtain multi-layer wavelet packet coefficient data; Performing energy density calculation processing on the multi-layer wavelet packet coefficient data to obtain time-frequency energy density data; Vectorized splicing processing is performed on the comprehensive gradient modulus data, the time-frequency energy density data, and the target gas concentration data to obtain the joint vector.

[0076] The fault classification module 24 is used to perform fault classification processing on the joint vector through a preset attribute graph attention network model to obtain a fault category label.

[0077] In an optional embodiment, the fault classification module 24 is specifically configured to: Performing node division processing on the joint vector to obtain an attribute vector of each monitoring point; Performing neighborhood relationship construction processing on the attribute vector according to preset heat conduction and mechanical connection rules to obtain edge weights between nodes; Calculate the attention coefficient according to the edge weights between the nodes to obtain the attention coefficient; The attribute graph attention network model performs weighted aggregation processing on the attribute vectors of the neighborhood nodes according to the attention coefficient to obtain the node feature vector; Softmax classification is performed on the node feature vector to obtain the fault category label.

[0078] The fault simulation module 25 is configured to perform multi-physics field coupling simulation processing according to the fault category label and the multi-source calibration data set to obtain a fault region coordinate set.

[0079] In an optional embodiment, the fault simulation module 25 is specifically configured to: Determining a corresponding fault physical field type according to the fault category label, and performing target physical field data screening processing on the multi-source calibration data set to obtain a target physical field data subset; Performing three-dimensional space grid division processing on the target physical field data subset to obtain a multi-physics field coupling simulation calculation grid; Performing thermal-electrical-mechanical-gas multi-field joint boundary condition assignment processing on the multi-physics field coupling simulation calculation grid to obtain a boundary constraint condition matrix; Performing finite element solution processing on the multi-physics field coupling simulation calculation grid according to the boundary constraint matrix to obtain fault response distribution field data; Performing isosurface extraction processing on the fault response distribution field data to obtain distribution information of high fault response areas; A spatial coordinate analysis process is performed based on the distribution information of the high fault response area to obtain the fault area coordinate set.

[0080] The life prediction module 26 is used to perform remaining life prediction processing based on the fault area coordinate set and preset historical degradation data through a preset dynamic Bayesian network model to obtain a remaining life probability distribution and maintenance instructions.

[0081] In an optional embodiment, the life prediction module 26 is specifically configured to: Performing spatial matching processing on the historical degradation data according to the fault area coordinate set to obtain target fault evolution data; Performing time-series segmented modeling processing on the target fault evolution data to obtain a multi-stage degradation state sequence; Performing state transition probability calculation processing according to the multi-stage degradation state sequence to obtain a time series state transition matrix; Performing latent variable joint distribution fitting processing on the time series state transfer matrix to obtain a dynamic Bayesian network prior parameter set; Performing latent variable joint distribution fitting processing on the time series state transfer matrix to obtain a dynamic Bayesian network prior parameter set; Performing decision threshold analysis on the remaining life probability distribution to obtain a maintenance risk assessment indicator; A maintenance instruction generation process is performed according to the maintenance risk assessment indicator to obtain the maintenance instruction.

[0082] It should be understood that the various variations and specific embodiments of the methods provided in the above embodiments are also applicable to the UHV GIS graphene-copper-based composite guide assembly fault diagnosis device of this embodiment. Through the above detailed description of the UHV GIS graphene-copper-based composite guide assembly fault diagnosis method, those skilled in the art can clearly understand the implementation method of the UHV GIS graphene-copper-based composite guide assembly fault diagnosis device in this embodiment. For the sake of brevity of the specification, it will not be described in detail here.

[0083] like Figure 3 , which is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0084] In a preferred embodiment of the present invention, the electronic device 3 may include, but is not limited to: a memory 31 , at least one processor 32 and at least one communication bus 33 .

[0085] Those skilled in the art should understand that Figure 3 The structure of the electronic device 3 shown does not constitute a limitation of the embodiment of the present invention. The electronic device 3 may also include more or less other hardware or software than shown in the figure, or a different component arrangement.

[0086] In some embodiments, the electronic device 3 is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits, programmable gate arrays, digital processors and embedded devices.

[0087] It should be noted that the electronic device 3 is only an example. Other existing or future electronic products that are suitable for this application should also be included in the scope of protection of this application and included here by reference.

[0088] In some embodiments, the memory 31 stores a computer program that, when executed by the at least one processor 32, implements all or part of the steps in the UHV GIS graphene-copper composite flow guide assembly fault diagnosis method. The memory 31 includes a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disk storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data. Furthermore, the computer-readable storage medium may primarily include a program storage area and a data storage area, wherein the program storage area may store an operating system, at least one application required for a function, and the like.

[0089] In some embodiments, the at least one processor 32 serves as the control core (Control Unit) of the electronic device 3. It connects the various components of the electronic device 3 using various interfaces and circuits. It executes programs or modules stored in the memory 31 and accesses data stored in the memory 31 to perform various functions and process data in the electronic device 3. For example, when executing the computer program stored in the memory 31, the at least one processor 32 implements all or part of the steps of the UHV GIS graphene-copper composite flow guide assembly fault diagnosis method described in the embodiments of this application; or implements all or part of the functions of the UHV GIS graphene-copper composite flow guide assembly fault diagnosis device. The at least one processor 32 can be comprised of an integrated circuit, such as a single packaged integrated circuit or multiple packaged integrated circuits with the same or different functions, including a combination of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips.

[0090] In some embodiments, the at least one communication bus 33 is configured to enable communication between the memory 31 and the at least one processor 32. Although not shown, the electronic device 3 may also include a power supply (e.g., a battery) to power various components. Preferably, the power supply can be logically connected to the at least one processor 32 via a power management device, thereby enabling the power management device to manage charging, discharging, and power consumption. The power supply may also include one or more DC or AC power supplies, a recharging device, a power failure detection circuit, a power converter or inverter, a power status indicator, and other components. The electronic device 3 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be detailed here.

[0091] The above-mentioned integrated unit implemented in the form of a software functional module can be stored in a computer-readable storage medium. The above-mentioned software functional module stored in a storage medium includes a number of instructions for causing an electronic device (which can be a personal computer, electronic device, or network device, etc.) or a processor to execute portions of the methods described in various embodiments of the present application.

[0092] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the module division is only a logical function division, and other division methods may be used in actual implementation.

[0093] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, and may be located in one place or distributed across multiple network elements. Some or all of the modules may be selected to achieve the purpose of this embodiment based on actual needs.

[0094] The above are all preferred embodiments of the present application, and are not intended to limit the scope of protection of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A method for diagnosing faults in a UHV GIS graphene-copper composite flow guide assembly, characterized in that: The method comprises: Perform dynamic calibration and anti-interference data acquisition and processing on the target diversion component to obtain a multi-source calibration data set; Performing a time-space-frequency domain fusion process on the multi-source calibration dataset by a preset Tucker decomposition method to obtain a cross-modal coupling feature set; Performing time-frequency-spatial domain joint feature extraction and encoding processing on the cross-modal coupling feature set to obtain a joint vector; Performing fault classification processing on the joint vector through a preset attribute graph attention network model to obtain a fault category label; Performing multi-physics field coupling simulation processing according to the fault category label and the multi-source calibration data set to obtain a fault area coordinate set; The remaining life prediction process is performed based on the fault area coordinate set and preset historical degradation data through a preset dynamic Bayesian network model to obtain the remaining life probability distribution and maintenance instructions.

2. The fault diagnosis method for the UHV GIS graphene-copper-based composite guide assembly according to claim 1 is characterized in that: The dynamic calibration of the target diversion component and the anti-interference data acquisition and processing to obtain a multi-source calibration data set include: Performing infrared thermal imaging on the target guide component to obtain surface grayscale values, and performing nonlinear temperature calibration on the surface grayscale values ​​to obtain calibrated temperature field data; Acquiring an acoustic emission signal from the target flow guide component to obtain an original acoustic emission signal, and performing frequency domain inverse filtering compensation processing on the original acoustic emission signal to obtain compensated optimized acoustic emission data; Performing microstrain acquisition on the target flow guide component to obtain measured strain data, and performing thermal expansion correction processing on the measured strain data according to the calibration temperature field data to obtain stress field data; Performing a partial discharge test on the target flow guide component to obtain an original discharge pulse signal, and performing time domain synchronous compensation processing on the original discharge pulse signal to obtain time domain optimized discharge pulse phase data; Performing insulation gas decomposition mass spectrum peak data collection and linear calibration processing on the target flow guide component to obtain target gas concentration data; Data fusion processing is performed on the calibration temperature field data, the compensated optimized acoustic emission data, the stress field data, the time domain optimized discharge pulse phase data and the target gas concentration data to obtain the multi-source calibration data set.

3. The fault diagnosis method for the UHV GIS graphene-copper-based composite flow guide assembly according to claim 2 is characterized in that: The performing time-space-frequency fusion processing on the multi-source calibration dataset by a preset Tucker decomposition method to obtain a cross-modal coupling feature set includes: Performing unified spatiotemporal mapping processing on the calibrated temperature field data, the compensated optimized acoustic emission data, the stress field data, and the time-domain optimized discharge pulse phase data to obtain a four-dimensional heterogeneous tensor; Factor decomposition is performed on the four-dimensional heterogeneous tensor using a preset Tucker decomposition method to obtain an original core tensor and a factor matrix set; Performing energy truncation processing on the original core tensor according to a preset energy proportion threshold to obtain a compressed feature tensor; A cross-modal coupling feature fusion storage process is performed on the compressed feature tensor and the factor matrix set to obtain the cross-modal coupling feature set.

4. The fault diagnosis method for the UHV GIS graphene-copper-based composite flow guide assembly according to claim 2 is characterized in that: The performing time-frequency-spatial domain joint feature extraction and encoding processing on the cross-modal coupling feature set to obtain a joint vector includes: Performing spatial mode reconstruction processing on the cross-modal coupling feature set to obtain gradient field data of each modality; Performing gradient modulus calculation processing on each modal gradient field data to obtain comprehensive gradient modulus data; Performing wavelet packet decomposition processing on the compensated optimized acoustic emission data to obtain multi-layer wavelet packet coefficient data; Performing energy density calculation processing on the multi-layer wavelet packet coefficient data to obtain time-frequency energy density data; Vectorized splicing processing is performed on the comprehensive gradient modulus data, the time-frequency energy density data, and the target gas concentration data to obtain the joint vector.

5. The fault diagnosis method for the UHV GIS graphene-copper-based composite guide assembly according to claim 1 is characterized in that: The performing fault classification processing on the joint vector by using a preset attribute graph attention network model to obtain a fault category label includes: Performing node division processing on the joint vector to obtain an attribute vector of each monitoring point; Performing neighborhood relationship construction processing on the attribute vector according to preset heat conduction and mechanical connection rules to obtain edge weights between nodes; Calculate the attention coefficient according to the edge weights between the nodes to obtain the attention coefficient; The attribute graph attention network model performs weighted aggregation processing on the attribute vectors of the neighborhood nodes according to the attention coefficient to obtain the node feature vector; Softmax classification is performed on the node feature vector to obtain the fault category label.

6. The fault diagnosis method for the UHV GIS graphene-copper-based composite flow guide assembly according to claim 1, characterized in that: The performing multi-physics field coupling simulation processing according to the fault category label and the multi-source calibration data set to obtain a fault area coordinate set includes: Determining a corresponding fault physical field type according to the fault category label, and performing target physical field data screening processing on the multi-source calibration data set to obtain a target physical field data subset; Performing three-dimensional space grid division processing on the target physical field data subset to obtain a multi-physics field coupling simulation calculation grid; Performing thermal-electrical-mechanical-gas multi-field joint boundary condition assignment processing on the multi-physics field coupling simulation calculation grid to obtain a boundary constraint condition matrix; Performing finite element solution processing on the multi-physics field coupling simulation calculation grid according to the boundary constraint matrix to obtain fault response distribution field data; Performing isosurface extraction processing on the fault response distribution field data to obtain distribution information of high fault response areas; A spatial coordinate analysis process is performed based on the distribution information of the high fault response area to obtain the fault area coordinate set.

7. The fault diagnosis method for the UHV GIS graphene-copper-based composite flow guide assembly according to claim 1 is characterized in that: The method of performing a remaining life prediction process based on the fault area coordinate set and preset historical degradation data using a preset dynamic Bayesian network model to obtain a remaining life probability distribution and maintenance instructions includes: Performing spatial matching processing on the historical degradation data according to the fault area coordinate set to obtain target fault evolution data; Performing time-series segmented modeling processing on the target fault evolution data to obtain a multi-stage degradation state sequence; Performing state transition probability calculation processing according to the multi-stage degradation state sequence to obtain a time series state transition matrix; Performing latent variable joint distribution fitting processing on the time series state transfer matrix to obtain a dynamic Bayesian network prior parameter set; Performing latent variable joint distribution fitting processing on the time series state transfer matrix to obtain a dynamic Bayesian network prior parameter set; Performing decision threshold analysis on the remaining life probability distribution to obtain a maintenance risk assessment indicator; A maintenance instruction generation process is performed according to the maintenance risk assessment indicator to obtain the maintenance instruction.

8. A fault diagnosis device for a UHV GIS graphene-copper composite guide assembly, characterized in that: The device comprises: The data acquisition module is used to dynamically calibrate the target diversion component and perform anti-interference data acquisition and processing to obtain a multi-source calibration data set; A data fusion module is used to perform time-space-frequency fusion processing on the multi-source calibration data set through a preset Tucker decomposition method to obtain a cross-modal coupling feature set; a joint extraction module, configured to perform time-frequency-spatial domain joint feature extraction and encoding processing on the cross-modal coupling feature set to obtain a joint vector; A fault classification module is used to perform fault classification processing on the joint vector through a preset attribute graph attention network model to obtain a fault category label; a fault simulation module, configured to perform multi-physics field coupling simulation processing according to the fault category label and the multi-source calibration data set to obtain a fault area coordinate set; The life prediction module is used to perform remaining life prediction processing based on the fault area coordinate set and preset historical degradation data through a preset dynamic Bayesian network model to obtain remaining life probability distribution and maintenance instructions.

9. An electronic device, characterized in that: The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for diagnosing faults of ultra-high voltage GIS graphene-copper-based composite guide components according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for diagnosing faults of ultra-high voltage GIS graphene-copper-based composite guide components according to any one of claims 1 to 7 are implemented.

Images