Partial discharge high-precision detection method and system for totally-enclosed GIS equipment

Abstract

The invention relates to the technical field of discharge detection, in particular to a high-precision detection method and device for partial discharge of totally-enclosed GIS equipment. The method comprises the following steps: carrying out metal shell three-dimensional structure modeling on the GIS equipment, and carrying out electromagnetic shielding evaluation so as to obtain a multi-point electromagnetic shielding evaluation value; acquiring an original discharge signal of each sampling point based on the electromagnetic wave sensor, and performing interference suppression processing and back propagation attenuation compensation according to a multi-point electromagnetic shielding evaluation value, so as to obtain an attenuation compensation discharge signal waveform; micro-vibration response parameters are collected based on an MEMS micro-vibration sensor, actual influence waveform point detection marking and physical vibration interference filtering are carried out on attenuation compensation discharge signal waveforms, and therefore pure optimized discharge signal waveforms are obtained. According to the invention, through efficient and accurate partial discharge detection, real-time early warning of the state of the GIS equipment is realized, and the stability and safety of the equipment are improved.

Description

Technical Field

[0001] The present invention relates to the field of discharge detection technology, and in particular to a high-precision detection method and system for partial discharge of fully enclosed GIS equipment. Background Art

[0002] With the rapid development and intelligentization of power systems, gas-insulated switchgear (GIS), a key component of high-voltage transmission systems, has been widely used in urban power grids, substations, and critical industrial facilities due to its compact structure, stable operation, and easy maintenance. Particularly under high voltage levels and complex operating conditions, fully enclosed GIS equipment, with its excellent insulation performance and superior protection capabilities, has become a core component of modern power system safety and stability. However, with extended operation and the added influence of external environments, aging and accumulation of defects in the insulation materials within GIS equipment are inevitable. Partial discharge (PD) is a significant precursor to insulation degradation and potential equipment failure.

[0003] Partial discharge (PD), a tiny discharge phenomenon caused by localized electrical breakdown in an insulation system, presents significant challenges to detection due to its weak and complex signal characteristics. Traditional PD detection methods rely on single sensing technologies such as current, voltage, or ultrasound, and are often applied in open or semi-open equipment environments. They struggle to adapt to the complex electromagnetic propagation characteristics within fully enclosed GIS equipment. Furthermore, due to the diverse and intense electromagnetic interference within GIS, traditional detection methods are significantly limited in accurately capturing and locating signals. This results in a high rate of misjudgment of discharge signals, making high-precision, real-time monitoring and fault warning impossible.

[0004] With the advancement of information technology and sensor technology, achieving highly sensitive and accurate partial discharge detection within fully enclosed structures has become a key technical challenge in ensuring the safe operation of GIS equipment. Existing partial discharge detection methods generally lack the ability to effectively separate and identify discharge signals in complex electromagnetic environments. Furthermore, they lack the accuracy to locate multiple, dynamically changing discharge sources, making them unable to meet the stringent requirements of modern smart grids for equipment safety diagnosis and early warning. Furthermore, the complex spatiotemporal evolution of partial discharge signals makes it difficult for traditional methods to reflect the changing trends of insulation conditions in real time, hindering the scientific nature of fault prediction and maintenance decisions. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention proposes a high-precision detection method and system for partial discharge of fully enclosed GIS equipment to solve at least one of the above technical problems.

[0006] To achieve the above objectives, the present invention provides a high-precision detection method for partial discharge in a fully enclosed GIS device. The GIS device is equipped with multiple electromagnetic wave sensor sampling points and MEMS microseismic sensors on its surface, including the following steps: Step S1: performing three-dimensional structural modeling of the metal shell of the GIS equipment and performing electromagnetic shielding evaluation to obtain multi-point electromagnetic shielding evaluation values; Step S2: collecting the original discharge signal of each sampling point based on the electromagnetic wave sensor, and performing interference suppression processing and reverse propagation attenuation compensation according to the multi-point electromagnetic shielding evaluation value, thereby obtaining an attenuation-compensated discharge signal waveform; Step S3: collecting micro-vibration response parameters based on the MEMS micro-seismic sensor, and performing actual influencing waveform point detection and marking on the attenuation compensation discharge signal waveform and filtering physical vibration interference, thereby obtaining a pure optimized discharge signal waveform; Step S4: Analyze the partial discharge propagation path based on the pure optimized discharge signal waveform, perform fuzzy positioning of the discharge area, and construct a probability distribution heat map of the discharge source area; Step S5: Perform multi-point synchronous radio frequency positioning analysis on the pure optimized discharge signal waveform, and perform reverse discharge simulation on the probability distribution heat map of the discharge source area to construct an accurate discharge signal distribution trajectory; Step S6: Evolve the discharge trend of multiple locations based on the precise discharge signal distribution trajectory, predict the insulation degradation situation, and construct a real-time discharge evolution warning image.

[0007] In this specification, a high-precision partial discharge detection system for a fully enclosed GIS device is provided, which is used to perform the high-precision partial discharge detection method for a fully enclosed GIS device as described above, including: An electromagnetic shielding evaluation module is used to perform three-dimensional structural modeling of the metal shell of the GIS equipment and perform electromagnetic shielding evaluation to obtain multi-point electromagnetic shielding evaluation values; An attenuation compensation module is used to collect the original discharge signal of each sampling point based on the electromagnetic wave sensor, and perform interference suppression processing and reverse propagation attenuation compensation according to the multi-point electromagnetic shielding evaluation value, so as to obtain an attenuation compensated discharge signal waveform; The physical vibration filtering module is used to collect micro-vibration response parameters based on the MEMS micro-seismic sensor, and to detect and mark the actual influencing waveform points and filter the physical vibration interference on the attenuation compensation discharge signal waveform, thereby obtaining a pure and optimized discharge signal waveform; The propagation path analysis module is used to analyze the partial discharge propagation path based on the pure optimized discharge signal waveform, perform fuzzy positioning of the discharge area, and construct a probability distribution heat map of the discharge source area; The reverse discharge simulation module is used to perform multi-point synchronous RF positioning analysis on the pure and optimized discharge signal waveform, and to perform reverse discharge simulation on the probability distribution heat map of the discharge source area to construct a precise discharge signal distribution trajectory; The degradation trend prediction module is used to analyze the discharge trend evolution of multiple locations based on the precise discharge signal distribution trajectory, predict the insulation degradation trend, and construct a real-time discharge evolution warning image.

[0008] The beneficial effects of the present invention are as follows: By introducing a three-dimensional structural modeling and multi-point electromagnetic shielding assessment mechanism based on metal casings, it effectively addresses the signal misjudgment and missed detection issues in traditional GIS partial discharge detection, which are caused by complex structures, strong shielding effects, and unclear electromagnetic wave propagation paths. The system uses high-precision finite element simulation to establish an electromagnetic propagation model, quantitatively analyzing the electromagnetic shielding capabilities of various areas within the device, thereby providing a precise basis for subsequent interference suppression and attenuation compensation of discharge signals. Based on this shielding assessment value, combined with the original discharge signal collected by the electromagnetic wave sensor, a reverse propagation model is constructed to perform attenuation compensation and interference filtering on the electrical signal, effectively enhancing the observability and recognition of weak discharge signals and improving the system's response sensitivity to low-intensity discharge events.

[0009] The system further incorporates MEMS microseismic sensors to achieve a synchronous detection mechanism with electromagnetic signals. By identifying abnormal disturbance patterns in the microvibration response signal, it accurately identifies non-electrical discharge waveform interference caused by physical vibrations (such as structural resonance and cable sway). Combined with waveform feature analysis techniques, it filters out vibration interference points to obtain a purer discharge signal. This multimodal sensing fusion strategy removes physical interference signals from the signal source, effectively improving the signal's structural integrity and diagnostic usability, and providing a more reliable foundation for identifying discharge propagation paths.

[0010] Based on the pure waveform, the system further performs discharge path tracing and fuzzy location analysis. Combining propagation delay information with a three-dimensional structural model, it constructs a spatial trajectory for discharge propagation and generates a heat map of the discharge source's regional probability distribution. This heat map, by overlaying parameters such as electric field intensity distribution and the shortest propagation probability, enables fuzzy spatial inference from "discharge presence" to "discharge location," providing visual and structured regional guidance for subsequent precise location and maintenance. Incorporating a multi-point RF synchronous sampling mechanism, the system achieves spatial reconstruction and trajectory inversion of discharge events. Through high-precision time delay difference calculation and inverse modeling algorithms, the system restores the true propagation path of the discharge source within the structure, constructing a high-resolution discharge distribution trajectory map, achieving millimeter-level discharge source location accuracy and providing a complete trajectory chain for discharge behavior modeling. Based on this precise discharge trajectory information, the system further introduces a discharge trend evolution analysis module, integrating time series modeling with structural health state deduction methods to dynamically assess the degradation state and evolution trends of multiple discharge sources.

[0011] By establishing a dynamic insulation degradation trend model, a quantitative assessment of insulation material aging, energy accumulation, and discharge frequency in key areas is achieved, creating a real-time discharge evolution warning image. This image not only reflects the current discharge state but also predicts future evolution direction and risk level, significantly enhancing operations and maintenance personnel's visual perception of equipment status and their ability to respond to warnings. By integrating multi-physics modeling, signal processing enhancement, multimodal data collaboration, discharge behavior trajectory reconstruction, and predictive warning technologies, this system overcomes the limitations of traditional GIS partial discharge detection, which relies on empirical judgment, is susceptible to interference, and has low positioning accuracy. This system achieves a technological leap from "detection" to "understanding," and from "static response" to "dynamic prediction" of internal GIS discharge events. The system possesses adaptive recognition capabilities, structured visual output, and real-time warning capabilities. It can be widely used in online monitoring, warning management, and predictive maintenance scenarios for GIS equipment in power systems, helping to transform and upgrade power equipment from condition monitoring to intelligent operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a schematic flow chart of the steps of a high-precision detection method for partial discharge of a fully enclosed GIS device according to the present invention; Figure 2 Detailed implementation flow chart of step S1; Figure 3 Detailed implementation flow chart of step S2; Figure 4 Schematic diagram of the detailed implementation steps of step S3. DETAILED DESCRIPTION

[0013] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0014] This application provides a high-precision method and system for detecting partial discharge in fully enclosed GIS equipment. The execution entities of this method and system include, but are not limited to, the following: mechanical equipment, data processing platforms, cloud server nodes, network upload devices, etc., which can be considered general computing nodes of this application. The data processing platform includes, but is not limited to, at least one of an audio and image management system, an information management system, and a cloud-based data management system.

[0015] See also Figures 1 to 4 The present invention provides a high-precision detection method for partial discharge of a fully enclosed GIS device, comprising the following steps: Step S1: performing three-dimensional structural modeling of the metal shell of the GIS equipment and performing electromagnetic shielding evaluation to obtain multi-point electromagnetic shielding evaluation values; Step S2: collecting the original discharge signal of each sampling point based on the electromagnetic wave sensor, and performing interference suppression processing and reverse propagation attenuation compensation according to the multi-point electromagnetic shielding evaluation value, thereby obtaining an attenuation-compensated discharge signal waveform; Step S3: collecting micro-vibration response parameters based on the MEMS micro-seismic sensor, and performing actual influencing waveform point detection and marking on the attenuation compensation discharge signal waveform and filtering physical vibration interference, thereby obtaining a pure optimized discharge signal waveform; Step S4: Analyze the partial discharge propagation path based on the pure optimized discharge signal waveform, perform fuzzy positioning of the discharge area, and construct a probability distribution heat map of the discharge source area; Step S5: Perform multi-point synchronous radio frequency positioning analysis on the pure optimized discharge signal waveform, and perform reverse discharge simulation on the probability distribution heat map of the discharge source area to construct an accurate discharge signal distribution trajectory; Step S6: Evolve the discharge trend of multiple locations based on the precise discharge signal distribution trajectory, predict the insulation degradation situation, and construct a real-time discharge evolution warning image.

[0016] In the embodiment of the present invention, see Figure 1 , is a schematic flow chart of the steps of a high-precision detection method for partial discharge of a fully enclosed GIS device according to the present invention. In this example, the steps of the high-precision detection method for partial discharge of a fully enclosed GIS device include: Step S1: performing three-dimensional structural modeling of the metal shell of the GIS equipment and performing electromagnetic shielding evaluation to obtain multi-point electromagnetic shielding evaluation values; In this embodiment, due to the extremely strong electromagnetic shielding capabilities of the metal casing of high-voltage GIS (Gas Insulated Switchgear) equipment, the electromagnetic waves generated by partial discharge (PD) experience significant attenuation, reflection, and path distortion during propagation. Therefore, to accurately analyze the propagation characteristics of PD signals, it is necessary to first establish a three-dimensional (3D) metal casing model that closely matches the actual structure. Based on this model, multi-point electromagnetic shielding simulation and evaluation are performed to obtain a spatial shielding strength distribution map. Based on the actual structural parameters of the on-site GIS equipment, a detailed model of the metal enclosure is constructed using 3D modeling software (such as SolidWorks or Autodesk Inventor). The modeling targets key components, including the busbar compartment, circuit breaker compartment, voltage transformer, grounding switch, and bushing connection section. Dimensional accuracy is required to be controlled within ±0.5 mm. Particular attention is paid to the precise modeling of details such as joints, flange connections, and grounding points, as these areas are often vulnerable to electromagnetic leakage. After modeling is completed, the model is exported to an electromagnetic simulation platform, such as ANSYS HFSS or CSTStudio Suite, for meshing and physical modeling. After entering the electromagnetic simulation stage, in order to simulate the propagation and shielding characteristics of high-frequency electromagnetic waves generated by partial discharge in the metal shell, it is necessary to set relevant material parameters and boundary conditions. The metal shell is usually made of aluminum alloy or stainless steel, and its conductivity is set to σ = 3.5× S / m; the relative magnetic permeability μ is approximately 1, assuming a perfect electric conductor (PEC) boundary. Simultaneously, the internal air medium is modeled as a homogeneous medium with ε = 1, and the discharge source is simulated as a pulsed current source (frequency range 100 kHz to 200 MHz) embedded in a specific insulating crack. Electromagnetic field sampling points are arranged at several typical locations in the model, including the inner wall of the housing at different distances from the discharge source, near key connection points, and at structural corners (a total of 30 to 50 points). Parameters such as the electric field intensity E and magnetic field intensity H are collected at each point, and their ratios to the radiation field at the initial source point are calculated. The corresponding shielding efficiency (SE) is then calculated: SE = 20· (E_0 / E_1) (unit: dB); where is the theoretical unshielded field strength, and E_1 is the actual measured field strength. Shielding efficiency greater than 80 dB is considered excellent, 60-80 dB is considered good, and below 60 dB is considered a weak area. After obtaining the shielding values ​​of all sampling points at different frequencies through simulation, a "multi-point electromagnetic shielding evaluation heat map" can be plotted to display the spatial distribution of electromagnetic shielding within the GIS enclosure. In a typical experimental environment, a 252 kV GIS model was tested with a simulation frequency range of 10 MHz to 150 MHz. The results showed that the shielding efficiency of most areas of the cabin was above 85 dB, but there were multiple weak shielding areas of 60-65 dB near the busbar and arrester connection flanges, requiring enhanced filtering and compensation processing in subsequent signal detection.

[0017] Step S2: collecting the original discharge signal of each sampling point based on the electromagnetic wave sensor, and performing interference suppression processing and reverse propagation attenuation compensation according to the multi-point electromagnetic shielding evaluation value, thereby obtaining an attenuation-compensated discharge signal waveform; In this embodiment, high-frequency electromagnetic wave sensors deployed at key points within the metal casing of GIS equipment collect high-frequency pulse signals released during partial discharge in real time. These sensors typically employ non-contact magnetic ring inductance sensors (such as HFCT) or adhesive-type ultra-wideband electric field sensors with a bandwidth range of 10 MHz to 300 MHz. These sensors are deployed based on the electromagnetic shielding characteristics of different areas of the GIS casing to capture the overall spatial distribution of electromagnetic signals. For a 220 kV GIS device, 12 electromagnetic sampling points were set at locations such as the bushing connection flange, busbar connection section, circuit breaker outlet, and voltage transformer contacts. The sensor sampling rate was set to 1 GS / s, with a 2 ms acquisition window. A high-pass filter was used to initially filter out low-frequency background noise (such as 50 Hz power frequency interference). The raw waveform signal exhibits large amplitude variations at multiple points, inconsistent phase delays, and localized envelope distortion. This is primarily due to a combination of structural shielding, propagation path loss, and metal reflection interference. Using the multi-point electromagnetic shielding evaluation value (SE distribution) obtained in the previous step S1, the electromagnetic shielding strength at each sampling point is used as a parameter and introduced into the signal processing model. For areas where shielding effectiveness is below 70 dB, the sensor signal is often accompanied by superimposed interference from ambient electromagnetic radiation. Therefore, an adaptive spectral subtraction algorithm based on spectral energy distribution is used to suppress the mid- and high-frequency bands of these signals while preserving the main component waveform of the discharge within the main frequency domain (typically 40 MHz to 150 MHz). To further eliminate multipath interference caused by multiple reflections from the shell's inner wall, a delay-deconvolution filtering method is used. Based on a reflection model, the arrival time of the interference waveform is estimated and removed from the original signal, thereby improving the clarity of the signal's main lobe. Experiments have shown that after interference suppression, the signal-to-noise ratio (SNR) improves by an average of 12 to 18 dB, and the recognition rate of the main discharge peak increases from 81% to over 94%. After interference suppression is complete, energy losses along the signal propagation path must be compensated based on the electromagnetic shielding model to restore the true waveform close to the original discharge source. This process is accomplished by establishing a propagation path attenuation model. The model is constructed based on the following parameters: Signal propagation path length (measured based on a three-dimensional structural model, in units of 0.8 to 3.5 m); Frequency-dependent energy attenuation coefficient of the shielded area (in dB / m); Metal surface reflection loss (estimate the number of reflections and reflection loss); Propagation medium properties in air / insulating gas (e.g. SF dielectric constant ε ≈ 1.001) Based on the above parameters, a frequency response function H(f) is constructed for each point. Deconvolution technology is used to perform frequency domain compensation, resulting in a frequency gain function G(f) = 1 / H(f). Finally, the original signal is subjected to compensation filtering to restore its amplitude and phase characteristics at the discharge source. The compensation process significantly improves the high-frequency signal (>100 MHz), significantly enhancing the clarity of the short pulse envelope. After interference suppression and attenuation compensation, the signal significantly reduces background electromagnetic interference, metal reflection multipath errors, and path distortion caused by the structure. This "attenuation-compensated discharge signal waveform" retains the true amplitude fluctuation, envelope variation, and microsecond-level phase characteristics of the discharge pulse, providing high-quality basic data for subsequent steps such as partial discharge path modeling, spatial positioning, neural network input, and twin mapping.

[0018] Step S3: collecting micro-vibration response parameters based on the MEMS micro-seismic sensor, and performing actual influencing waveform point detection and marking on the attenuation compensation discharge signal waveform and filtering physical vibration interference, thereby obtaining a pure optimized discharge signal waveform; In this embodiment, in fully enclosed GIS equipment, partial discharges, in addition to generating electromagnetic waves, can also induce mechanical microseismicity of a certain amplitude. This vibration response is particularly pronounced during partial discharges caused by casing defects, electrode tip discharges, and poor contact. However, during equipment operation, numerous non-electrical sources may exist, such as circuit breaker operation, external impacts, and structural resonance. These interferences can indirectly affect electromagnetic signal acquisition, resulting in spurious discharge waveforms. Therefore, it is necessary to deploy MEMS (Micro-Electro-Mechanical Systems) microseismic sensors to collect local microseismic responses and perform time-series mapping with the electrical signal waveform to identify and eliminate physical interference points, ultimately optimizing and purifying the partial discharge signal. The MEMS sensors are high-sensitivity triaxial accelerometers (such as the ADXL355) with a minimum resolution of 25 µg, a bandwidth set to 0.1 Hz to 2 kHz, and a response time of less than 1 ms. The sensors are deployed at mechanical connection points adjacent to the electromagnetic wave sensors, including the bracket connection, grounding flange, housing weld area, and the circuit breaker housing. A total of 10 microseismic sampling points are deployed. The sampling frequency was set at 10 kHz, and vibration data was continuously collected for 50 ms before and after each partial discharge event to ensure coverage of the complete mechanical response window. Fast Fourier transform (FFT) and short-time energy analysis were used to extract the dominant frequencies (5 Hz–500 Hz) and vibration energy distribution of the microseismic signals. Mechanical vibrations corresponding to real partial discharge events typically exhibit the following characteristics: a response delay between 0.53 ms, a duration less than 10 ms, and a vibration envelope curve exhibiting high-frequency attenuation. Structural interference, on the other hand, is often characterized by sustained oscillations with a frequency distribution below 50 Hz. The "attenuation-compensated discharge signal waveform" obtained in step S2 was jointly analyzed with the microseismic response data for the corresponding time period to construct an "electromagnetic-seismic temporal alignment matrix." A sliding time window cross-correlation function was used to quantify the time delay relationship between the signals and identify the temporal coupling points between the main peak of the electromagnetic wave and the microseismic response. Waveform segments where there is strong temporal coupling between the vibration response and electromagnetic signal changes (R>0.75) are marked with a "vibration correlation marker" within the waveform, and the marker point is highlighted within a ±1 ms range. This process helps identify secondary electromagnetic disturbance waveforms caused by physical structural vibration, such as continuous low-amplitude envelope oscillations caused by loose bushings, mid-frequency pseudo-bands caused by cable joint vibration, and repetitive pulses induced by metal shell resonance. These marker points will be identified as key targets in the subsequent physical interference filtering process.

[0019] Based on vibration marker information, a multidimensional interference identification vector is constructed, encompassing five metrics: time location, vibration amplitude, duration, frequency center, and phase fluctuation. Using the K-means clustering method, all electromagnetic waveform segments are divided into "high-vibration coupling segments" and "low-vibration coupling segments," with the former being suspected physical interference signals. For high-vibration coupling segments, physical interference is filtered out using a time-domain envelope filter and a harmonic component removal algorithm. Specifically, to address the pulse stacking phenomenon caused by vibration, wavelet denoising techniques (such as the Daubechies-4 wavelet) are used to extract the main lobe of the main discharge waveform and remove asymmetric tail waves and pseudo-edge responses. This processing preserves the main components of the discharge waveform while significantly suppressing non-discharge components caused by vibration. Experimental results show that after physical interference removal, the false detection rate (misidentification of non-discharge signals as PD) in the signal decreases by 78%, effectively improving the signal purity of the partial discharge waveform.

[0020] Step S4: Analyze the partial discharge propagation path based on the pure optimized discharge signal waveform, perform fuzzy positioning of the discharge area, and construct a probability distribution heat map of the discharge source area; In this implementation, based on pure discharge signals collected synchronously by multiple sensors (such as electromagnetic wave sensors deployed at different nodes in the GIS), statistical feature extraction combined with deep learning structural modeling analyzes the signal propagation patterns and identifies the possible spatial locations of discharge sources within the device. For multiple spatial sampling points (typically 12 to 16), the system captures the electromagnetic waveform collected for each discharge event within a unified time window (for example, a 50 μs window) and extracts the following key features for spatial propagation modeling: Signal arrival time T(arrival): determined based on the threshold starting point and the first spike detection; Peak amplitude (A(max)): measures the spatial decay of discharge intensity; Energy envelope (E(env)): used to reflect the persistence characteristics of the waveform; Frequency distribution (F(spectrum)): Analyze the spectrum characteristics through fast Fourier transform; Signal phase (φ): reflects the coherence information during the propagation process.

[0021] In the spatial dimension, the system constructs a network model of the discharge propagation path based on these characteristic differences. Using the fitted relationship between three-dimensional Euclidean distance and propagation delay, it reconstructs the propagation trajectory of the discharge signal within the metal shell and the SF6 gas medium. The path data is encoded in a graph structure, with each node corresponding to a sensing point and each edge representing a propagation path. The edge weight is the propagation time or energy decay.

[0022] Based on the propagation graph, the system introduces a pre-trained spatial graph convolutional neural network (GCN) and combines it with the self-attention mechanism to perform multi-path spatial correlation modeling. The specific structure is as follows: The first stage: The graph convolution layer aggregates the propagation path information between nodes and learns the potential spatial signal correlation relationship; Phase 2: The self-attention mechanism strengthens the perception of the directional characteristics of transmission by dynamically allocating weights to different paths; The third stage: The fully connected layer outputs the predicted probability of each spatial point as the discharge source.

[0023] The training data comes from a large number of historical experimental samples with known partial discharge source locations. The network learns spatial propagation patterns through end-to-end supervised training. The experimental dataset includes typical discharge types such as tip discharge, suspended potential discharge, and surface discharge, and the equipment structures cover 220 kV and 500 kV GIS models. Training accuracy exceeds 90%.

[0024] The model predicts the probability distribution of each region within the 3D GIS structure as a discharge source. The system maps this result into a 3D probability heat map. The GIS equipment space is discretized using voxel partitioning (sliced ​​at a 0.5-meter granularity), with each voxel containing a discharge probability value. Combining the equipment CAD model, geographic reference system, and electrical component layout, the system color-codes each voxel, creating an intuitive visualization interface.

[0025] Heatmap color coding goes from blue (low probability) → green → yellow → red (high probability); Support users to dynamically rotate and zoom the 3D structure of the device to observe the possible distribution of discharge sources; Areas with a probability higher than a set threshold (e.g., 0.7) are automatically marked as core areas of partial discharge risk.

[0026] This heat map is generated once for each partial discharge event, and the judgment threshold is dynamically adjusted based on the real-time operating status of the equipment and voltage level, ensuring the sensitivity and stability of the visualization. In actual testing, the system achieved an 89% accuracy rate in regional positioning on a dataset of known partial discharge sources, with an average spatial positioning error within 1.3 meters, meeting the acceptable standards for actual power operations and maintenance.

[0027] Step S5: Perform multi-point synchronous radio frequency positioning analysis on the pure optimized discharge signal waveform, and perform reverse discharge simulation on the probability distribution heat map of the discharge source area to construct an accurate discharge signal distribution trajectory; In this embodiment, for each discharge event, the system extracts the following key information from multiple sampling points: Signal arrival timestamp (T(arrival)): Based on the precise time point of the waveform rising edge, it combines threshold judgment and interpolation methods to achieve sub-sampling accuracy; RF signal waveform characteristics: such as the main band center frequency (approximately 100 MHz to 500 MHz), amplitude, and phase information; Timing phase difference: used to identify the propagation path and direction.

[0028] Time difference of arrival (TDoA) positioning technology, combined with three-dimensional spatial coordinates, calculates the propagation time difference between each sampling point. The RF signal source location is inferred based on the electromagnetic wave propagation speed within the device (approximately 0.6-0.8 times the speed of light, accounting for dielectric and metal reflection attenuation). Multi-point data fusion and weighted least squares (WLS) are used to estimate spatial position, ensuring positioning accuracy exceeding 1.5 meters. Using the heat map of the discharge source probability distribution generated in step S4, the system selects high-probability areas (the probability threshold is generally set above 0.7) within the heat map as candidate discharge source locations for simulation verification. The reverse discharge simulation utilizes electromagnetic field numerical simulation technology that combines the finite element method (FEM) with the finite-difference time-domain method (FDTD). The simulation model is based on the device's 3D CAD structure and is refined to a millimeter-level grid cell to ensure accurate reproduction of partial discharge characteristics.

[0029] The simulation process mainly includes: Discharge source excitation input: A waveform excitation signal is applied at the candidate position. The waveform parameters are consistent with the pure optimized discharge signal waveform, including pulse amplitude, spectral characteristics and duration (generally the pulse width is in the range of 10-50 ns); Electromagnetic wave propagation simulation: simulate the reflection, refraction, and attenuation process of discharge signals in the metal shell and insulating medium inside the GIS; Simulation output: The simulated waveform and energy distribution received at each sampling point are obtained. By comparing the correlations between the simulated waveforms and the actual waveforms (such as cross-correlation coefficients and peak time deviations), the system evaluates the plausibility of candidate discharge source locations. Optimization algorithms (such as genetic algorithms or particle swarm optimization) are used to adjust the discharge source location and waveform parameters, iteratively improving the consistency between simulation and measured data to achieve precise discharge source location. Based on the reverse simulation verification results, the system ultimately determines a set of high-confidence discharge source locations and corresponding signal propagation paths. Furthermore, incorporating a spatial gridding model, the discharge signal propagation path is refined into a continuous trajectory, depicting the detailed process of signal propagation from the source to surrounding equipment structures. This trajectory not only contains a sequence of spatial coordinates but also includes multi-dimensional information such as signal strength, spectral variations, and propagation delay. The trajectory data is used for subsequent partial discharge behavior analysis, fault mechanism research, and equipment maintenance decision support. In a typical 220 kV GIS equipment test environment, the discharge source location error achieved in this step was kept within 1.3 meters, and the signal propagation path fitting error was less than 5%, significantly outperforming traditional location methods.

[0030] Step S6: Evolve the discharge trend of multiple locations based on the precise discharge signal distribution trajectory, predict the insulation degradation situation, and construct a real-time discharge evolution warning image.

[0031] In this example, the discharge signal distribution trajectory data obtained through precise multi-sensor positioning is integrated to form a spatiotemporal continuous trajectory sequence for each discharge source. To ensure analysis accuracy, the trajectory data requires filtering and noise removal. Common methods include wavelet transform denoising or Kalman filtering. Filter parameters are set based on the actual experimental environment, such as a sampling frequency of 2 MHz and a filter cutoff frequency of 50 kHz to eliminate high-frequency environmental interference signals. The trajectory sequence is further interpolated to ensure continuity. By constructing a multidimensional time series model, the temporal variations in intensity, frequency, and spatial position of each discharge point are analyzed. A common method is time series analysis and trend detection based on a sliding window. Based on the experimental design, the sliding window length is set to 1 hour, with a window step of 10 minutes. The number of discharges, average discharge amplitude, and peak discharge amplitude within the window are counted to form a trend vector. Cluster analysis (such as the density-based DBSCAN algorithm) is then used to identify the dynamic evolution of the discharge source cluster, revealing the emergence of new discharge points, the spread of existing discharge points, and their migration trends. In the experimental parameters, the DBSCAN neighborhood radius ε was set to 0.05 meters, and the minimum sample number MinPts was set to 5 to ensure the stability of the clustering results. Based on the model's predictions, a discharge evolution warning image was generated in real time. The image uses a 2D or 3D GIS equipment structure diagram as a background, overlaid with the spatial location of the discharge point and risk level identifiers. A color gradient (green represents safety, yellow indicates warning, and red indicates alert) is used to display the discharge evolution. A dynamic effect uses animation to display the trajectory of the discharge source and the real-time change of risk level. The image generation cycle was set to update every 5 minutes to ensure the timeliness of the warning information. This image can be displayed on the GIS monitoring platform interface, allowing operators to intuitively understand the insulation status of the equipment and potential risks.

[0032] In this embodiment, refer to Figure 2 , is a flowchart of the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Performing a three-dimensional structural analysis on the metal shell of the GIS device to obtain the three-dimensional structural characteristics of the surface of the GIS device; Conduct 3D modeling based on the 3D structural features of the GIS equipment surface and construct a metal shell model; Identifying the material properties of the metal shell, simulating the electromagnetic wave propagation and reflection distribution inside the metal shell model, and constructing an electromagnetic reflection twin model of the metal shell; Perform electromagnetic signal distribution coupling on the electromagnetic reflection twin model of the metal shell to obtain an electromagnetic signal distribution map; Electromagnetic shielding evaluation is performed based on the electromagnetic signal distribution diagram to obtain multi-point electromagnetic shielding evaluation values.

[0033] In this example, the complete 3D surface features of the metal enclosure of a GIS (metal-enclosed switchgear) device are captured to provide basic geometric information for subsequent electromagnetic simulation and modeling. First, a 360-degree scan of the GIS device is performed using a non-contact 3D scanning device, such as a laser 3D scanner (e.g., the FARO Focus S350 or Leica RTC360). To improve scanning accuracy, segmented scanning combined with calibration point alignment is used to ensure an overall stitching error of no more than ±0.2 mm. The scanning resolution is set to ≤0.1 mm to ensure complete capture of surface details, particularly the geometric features of electromagnetically sensitive areas such as joints, bolt holes, and welds. The resulting point cloud data is imported into professional point cloud processing software (such as Geomagic Wrap or PolyWorks) for preprocessing, including noise point filtering, surface reconstruction, mesh optimization, and hole repair. Through surface reconstruction and topological analysis, a complete 3D geometric model of the GIS metal enclosure is obtained. This provides accurate boundary conditions for subsequent modeling and also extracts the position and dimensional parameters of structural elements (e.g., elbow segments, flanges, and grounding bolt connection points). This step not only provides geometric shape data but also records local curvature and thickness variations, serving as important parameter inputs for subsequent electromagnetic modeling. After obtaining high-precision 3D structural feature data, parametric modeling of the GIS equipment metal enclosure is performed using 3D modeling software (such as SolidWorks, CATIA, or ANSYS SpaceClaim). This modeling not only ensures a faithful reproduction of the GIS equipment's structural shape but also maintains consistency in physical properties such as material thickness and connection methods (e.g., screws or welds). During the modeling process, the scanned point cloud data is surface fitted and converted into CAD geometry. Volume modeling is then performed to construct a digital prototype of the complete enclosure structure. Specifically, the equipment is broken down into functional units (e.g., circuit breaker compartment, current transformer compartment, busbar compartment, etc.), with independent part models created for each unit to facilitate modular simulation. During this process, the thickness of the metal plates at different locations (e.g., 5 mm for steel plates and 8 mm for aluminum alloy enclosures) is accurately modeled, and structural details such as interfaces, fasteners, and flange connections are added. At the same time, for simulation needs, the positions of the electromagnetic signal injection port and the detection port are reserved in the modeling for setting boundary conditions and excitation sources in subsequent simulations. This modeling stage not only completes the establishment of the geometric structure, but also lays the foundation for subsequent material definition, electromagnetic property assignment, and boundary condition setting, ensuring that the entire system has physical response characteristics consistent with real GIS equipment. The material composition of the metal shell of the GIS equipment is determined through literature review and experimental verification, such as the common aluminum alloy 6061 or stainless steel 304, and its related electromagnetic parameters are obtained, such as electrical conductivity (σ), magnetic permeability (μ), relative dielectric constant (εr), etc. For example, the electrical conductivity of aluminum alloy 6061 is 3.77× S / m, relative magnetic permeability is 1, and dielectric constant is ~1.05.

[0034] Using a vector network analyzer (such as the Keysight N5227B), the metal plate's S-parameters are measured, and its complex dielectric constant is inferred from the actual reflection and transmission coefficients, further improving the realism of the simulation model. After entering the material parameters, an electromagnetic simulation model is established using electromagnetic simulation software (such as CST Microwave Studio or ANSYS HFSS). The boundary conditions are set to "perfect conductor boundary" or "open circuit boundary," and the excitation source is set to a 1 ns pulse current source or pulse voltage source (with an amplitude of 10 V). The propagation path and reflection characteristics of the partial discharge signal within the shell are simulated.

[0035] The simulations are solved using the finite-difference time-domain method (FDTD) or the finite element method (FEM), with a sampling accuracy of at least 20 points per wavelength to ensure accurate signal propagation within the 30 MHz–300 MHz frequency band. Post-processing extracts electromagnetic characteristics such as electric field strength (E-Field), magnetic field distribution (H-Field), reflection paths, and interference zones. This constructs an "electromagnetic reflection twin model" of the metal enclosure, effectively simulating the internal response characteristics of the device under electromagnetic wave excitation. The simulated multi-band electromagnetic reflection data is then coupled with actual signal detection points to construct a complete spatial distribution map of the electromagnetic signal. This process combines the simulated electromagnetic field data with the sensor layout (e.g., patch antennas, electric field probes, etc.) for multi-point collaborative analysis. Multiple detection points are first set (typically 10–20 key locations, such as near the ground wire, at the housing bend, and at the connection flange). Each point serves as an observation point, and E / H field strength data is extracted within different time windows. During the simulation, a sampling interval of 10 ps was set for each sampling point, and the total simulation time was 100 ns, capturing the complete discharge signal propagation process. The field distribution characteristics at different frequencies were obtained by transforming the time-domain response into the frequency-domain response. Using a data fusion algorithm written in MATLAB or Python, the electric field data collected at multiple points were fused using spatial interpolation (such as radial basis functions or kriging interpolation) to generate a two-dimensional or three-dimensional electromagnetic signal distribution map. This map not only reflects the main path of electromagnetic wave propagation but also shows areas of high reflection, signal focal points, and possible electromagnetic interference blind spots, providing an important basis for subsequent electromagnetic shielding evaluation. By analyzing this electromagnetic signal distribution map, the electromagnetic shielding performance of the metal casing at various key locations was evaluated. The evaluation metric used was "shielding effectiveness (SE)." The experimental simulation excitation source was a standard high-frequency pulse source (frequency range 50 MHz–1 GHz). Through combined simulation and experimental testing, shielding effectiveness was measured at different frequencies. The SE value at the corresponding frequency was calculated for each test point to construct a "frequency-domain shielding curve." The average and minimum SE values ​​at multiple frequencies were further extracted as comprehensive evaluation indicators. Experiments show that a well-performing shielding structure should have an SE value of at least 80 dB (at 500 MHz), while the SE value at localized weak points can drop as low as 40 dB. Combined with a heat map display of the distribution graph, areas of poor shielding performance can be identified, such as discontinuous welds, poor flange sealing, and loose fasteners. Based on this data, engineers can optimize the GIS equipment structure, such as by adding shielding sheets, improving sealing designs, or rerouting grounding lines, significantly improving the purity and accuracy of partial discharge detection signals. Multi-point electromagnetic shielding evaluation values ​​will serve as an important reference for determining the electromagnetic compatibility (EMC) of GIS equipment and the quality of the signal detection environment, ensuring the reliable operation of high-precision partial discharge detection systems.

[0036] In this embodiment, the specific steps of performing reverse propagation attenuation compensation on the interference suppression discharge signal curve to obtain the attenuation compensated discharge signal waveform are: Define multiple timestamps; Extracting the original discharge signal of each collection point at each time node according to the multiple time stamps to obtain the original discharge signal at multiple moments; Calculating the signal differences at corresponding moments based on the original discharge signals at multiple moments to obtain signal difference values ​​at different moments; The signal difference values ​​at different times are averaged to obtain the average value of the signal difference; The signal propagation attenuation of the metal casing of the GIS equipment is analyzed based on the average signal difference and the multi-point electromagnetic shielding evaluation value, thereby obtaining the propagation attenuation characteristics of the metal casing signal; Based on the propagation attenuation characteristic, reverse propagation attenuation compensation is performed on the interference suppression discharge signal curve, thereby obtaining an attenuation compensated discharge signal waveform.

[0037] In this embodiment, during the detection of partial discharge signals, the signal usually appears as a non-stationary pulse waveform with an extremely short duration (usually between tens and hundreds of nanoseconds) and a high frequency (500 MHz to 2 GHz). Therefore, a high sampling rate acquisition device must be used to accurately characterize its time characteristics. In order to perform a timing analysis of the propagation and change process of the signal, it is first necessary to discretize the waveform along the time axis and define a series of physically meaningful timestamps. A high-bandwidth digital oscilloscope (such as Keysight DSOS804A, 8 GHz bandwidth, 20 GS / s sampling rate) is used to sample the UHF partial discharge signal. In the acquired time domain waveform, a set of equidistant timestamps is set at fixed intervals (such as 5 ns). -T nAlternatively, based on the signal energy gradient, several non-uniform timestamps (e.g., T1=0 ns, T2=10 ns, T3=12 ns, T4=17 ns, etc.) can be adaptively set before and after the main discharge pulse to enhance the ability to analyze key signal turning points. The timestamp setting should balance timing coverage and analysis accuracy. It is generally recommended to fully cover the main waveform period (e.g., setting 40 timestamps within a 200 ns range), while ensuring that the interval between adjacent timestamps is less than 1 / 10 of the signal's main frequency period to facilitate effective feature extraction and difference calculation. GIS equipment is typically equipped with multiple UHF sensors (e.g., eight distributed across different compartments and the inner wall of the hull) to achieve synchronous spatial signal acquisition. Each sensor generates an independent discharge waveform, containing the superposition of the target signal and background interference. In this step, the signal values ​​at each set timestamp are extracted from the complete original waveform at each acquisition point to form a discrete time-domain signal set that can be used for timing comparison. For each sensor channel's raw data (e.g., a 20,000-point, 1 μs sampling sequence), the corresponding time position is searched according to a defined timestamp list. The amplitude at each time point is then obtained using linear or cubic spline interpolation. This signal collection not only records the energy distribution of the discharge event in space but also includes its propagation characteristics over time. Combined with the sensor layout diagram, this allows for preliminary inference of the electromagnetic wave's propagation path from the discharge source and the reflection / attenuation trends on various structural surfaces, laying the foundation for subsequent differential analysis and attenuation modeling. After obtaining the signal distribution at multiple timestamps, the next step is to assess the dynamics of the signal over time by calculating the signal differences between adjacent moments. This difference reflects the propagation velocity of the partial discharge signal within the metal casing, as well as the degree of reflection superposition and interference, and is a key parameter for analyzing propagation paths and shielding effects. Difference calculations employ vector dissimilarity metrics such as Euclidean distance or root mean square difference (RMS). This process can be further validated using principal component analysis (PCA) or time-frequency distribution plots. To extract the overall disturbance level during signal propagation, it is necessary to statistically average the signal differences over all time periods to obtain the Mean Signal Difference (MSD). This value reflects the average disturbance intensity caused by the structure during the overall propagation process and is one of the indicators for quantifying the electromagnetic complexity of the GIS metal shell.

[0038] Tests on a set of GIS discharge samples revealed MSD values ​​ranging from 2.1 mV to 4.8 mV. An MSD exceeding 5 mV indicates strong reflection from the shell structure, severely restricting electromagnetic signal propagation. This value serves as one of the fundamental input parameters for propagation attenuation modeling in system modeling. By combining the mean signal difference (MSD) with the shielding effectiveness (SE) of each acquisition point, a comprehensive attenuation model for electromagnetic wave propagation due to the GIS metal shell can be established. This model accounts for both multipath signal loss within the structure and the frequency-selective shielding effect created by the enclosed structure.

[0039] The attenuation index AI (Attenuation Index) can be calculated using the following relationship: AI = MSD / SE; Where SE is the average shielding value of all acquisition points (in dB), representing the attenuation level of the signal entering or passing through the area.

[0040] In experimental tests, if the MSD is 3.5 mV and the SE is 28 dB, the AI ​​is approximately 0.125 mV / dB. This value can be used to construct propagation attenuation maps for each structural region and visualize the signal attenuation trends in different structural regions through heat maps. For example, the AI ​​value for the straight pipe section within the shell is low (0.07-0.12), while the AI ​​value at the tee and angle joints is significantly higher (reaching above 0.25), indicating that structural shape is the primary factor affecting signal attenuation. To restore the true discharge signal waveform, the detected signal must be subjected to reverse propagation attenuation compensation. This process aims to eliminate signal energy attenuation and morphological distortion caused by the shell structure, making the original discharge signal closer to the intrinsic characteristics of the discharge source in the time and amplitude domains. The compensation model is constructed based on the aforementioned AI and propagation path information. The AI ​​is applied to the signal at each acquisition point to construct an inverse gain function, and the detected waveform is then weighted and restored. Compensation methods can use inverse filtering or wavelet multiscale reconstruction techniques to compensate for signal distortion in the frequency domain, avoiding amplification of high-frequency distortion. In experiments, the compensated signal showed a 12-28% increase in the main pulse amplitude, a steeper rising edge, and an upward shift in the main frequency component, significantly improving the accuracy of discharge source location and feature classification. This waveform can be fed into a deep neural network model to identify partial discharge types and locate fault areas, providing key technical support for the intelligent operation and maintenance of GIS equipment.

[0041] In this embodiment, refer to Figure 3 , is a flowchart of the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Based on multiple electromagnetic wave sensor sampling points, the original discharge signal of each sampling point is continuously collected; Performing a multi-scale wavelet transform on the original discharge signal to obtain time-frequency characteristics of the discharge signal at multiple scales; Performing time-frequency fluctuation analysis on each scale of the discharge signal time-frequency characteristics, and extracting the signal fluctuation curve of each scale; Identifying abnormal noise fluctuations on the signal fluctuation curve and marking interference source fluctuation points; Calculating the fluctuation intensity of the interference source fluctuation point, and performing interference suppression processing on the signal fluctuation curve to obtain an interference suppression discharge signal curve; The interference suppression discharge signal curve is subjected to reverse propagation attenuation compensation according to the multi-point electromagnetic shielding evaluation value, thereby obtaining an attenuation compensated discharge signal waveform.

[0042] In this embodiment, the GIS equipment has a complex internal structure, and partial discharge signals propagating within the enclosure are subject to multiple reflections, diffraction, and path differences. Therefore, a multi-point synchronous sampling strategy is required to capture signal variations at different locations. In practical deployments, 6 to 10 ultra-high-frequency (UHF) electromagnetic wave sensors are typically installed at key locations, such as circuit breakers, disconnectors, and busbar compartments, to form a comprehensive acquisition network. The acquisition equipment utilizes a high-bandwidth UHF receiver front-end module (frequency range 300 MHz–1.5 GHz) combined with a 20 GS / s high-speed oscilloscope (such as the Keysight UXR0204A) to achieve nanosecond sampling accuracy. The sampling time window is typically set between 500 ns and 1 μs to ensure complete capture of the entire discharge pulse. Each acquisition point acquires 20,000 raw discharge waveform data points per round. All sensors use a unified sampling timebase (GPS synchronization or fiber-optic triggering system) to ensure strict temporal consistency of signals at different sampling points, providing a foundation for subsequent signal comparison and frequency domain analysis. Raw discharge signals often exhibit nonstationary, transient pulse characteristics with complex time-frequency characteristics. Therefore, a multi-scale wavelet transform (MWT) is used for signal analysis. In experiments, commonly used wavelet basis functions include Daubechies wavelets (such as db4 or db6) and Symlets wavelets (such as sym5). These wavelet bases have excellent localization properties and are suitable for time-frequency analysis of discharge pulse signals. The raw signal is input into a wavelet decomposition module for continuous wavelet transform (CWT) or discrete wavelet transform (DWT). The multi-scale decomposition level is set, typically 6 to 8 levels, depending on the frequency range of the signal in the experiment. Each level of wavelet decomposition yields approximation coefficients (A) and detail coefficients (D). The detail coefficients correspond to high-frequency components and typically contain information about localized discharge pulses. In experiments, the frequency layers are controlled between 0.5 MHz and 1.5 GHz and distributed on a logarithmic scale, with scale s1 corresponding to 1.5 GHz, s2 to 750 MHz, and so on up to s8, corresponding to the fundamental frequency below 10 MHz. Wavelet transforms create a two-dimensional time-frequency energy spectrum, with time points on the horizontal axis and frequency or scale on the vertical axis. Sudden energy changes in the spectrum indicate possible moments of partial discharge events. This process significantly improves the separability of the discharge signal, providing a clear signal structure for subsequent scale fluctuation analysis and interference removal.

[0043] After obtaining the time-frequency characteristics of the signal at multiple scales, the trend of signal energy variation at each scale needs to be analyzed. This step primarily focuses on extracting the fluctuation curve of the signal intensity over time at each scale, namely, the energy envelope curve. Specifically, a sliding window energy calculation method is applied to the detail coefficients at each scale, setting the sliding window size to 64 points (approximately corresponding to 25 ns), sliding 32 points at a time. Local energy (i.e., the sum of squared coefficients) is calculated and output as a time series. In experiments, to improve the response to weak discharge signals, the energy curves at each scale are normalized to eliminate energy deviations between scales due to filter characteristics. The resulting curve is known as the "signal fluctuation curve," graphically represented as a series of amplitude peaks and valleys, corresponding to transient energy accumulations caused by local discharge events or background interference. Furthermore, statistical characteristics such as short-term variance and short-term kurtosis can be introduced to measure the degree of fluctuation within each time period. These fluctuation curves will be used later to identify abnormal interference fluctuation points, so sensitive capture of time-frequency fluctuations is crucial. What is ultimately obtained is a collection of fluctuation curves at multiple scales, each of which represents the change in discharge activity over time within a specific frequency bandwidth, laying the foundation for interference identification and signal recovery.

[0044] Due to the complex operating environment of GIS and the diverse sources of electromagnetic interference, such as wireless communication signals, cable RF leakage, and switching operations, these non-PD signals can appear as pulses in the fluctuation curve, affecting the accuracy of the judgment. This step uses statistical analysis and time-frequency template matching to identify anomalies in the fluctuation curve at each scale. First, a background noise baseline model is established for historical data to extract the average energy distribution and peak distribution range of the interference characteristics. For example, if a long-term fixed-frequency disturbance signal is found at scale s3 (approximately 375 MHz), it can be identified as a periodic communication interference source. A threshold discrimination method is introduced in the experiment, setting a dynamic energy threshold, such as the average energy value plus two standard deviations. Points exceeding this threshold are considered outliers. Furthermore, a signal duration constraint is introduced: if the fluctuation width exceeds the typical duration of PD (e.g., 0.5 μs), it is marked as an interference source. Synchronous fluctuation points appearing at multiple scales are also considered systematic interference signals. In the template matching phase, a similarity matching algorithm is constructed using known interference waveforms. The correlation coefficient of the current fluctuation point is compared, and those with a similarity above a threshold (e.g., 0.8) are marked. Finally, interference source fluctuation points are marked in the fluctuation curves at all scales for subsequent interference suppression algorithm processing. This step significantly improves the distinguishability between partial discharge and background interference signals. After the interference fluctuation points are marked, the intensity characteristics of the interference fluctuations need to be further quantified and suppressed from the signal. The fluctuation intensity is calculated using energy integration and signal-to-noise ratio evaluation. The former calculates the sum of the energy within a certain range near the abnormal fluctuation point to determine its interference contribution; the latter compares the discharge signal at that point with the average noise energy in the neighborhood to determine its suppression priority. The interference suppression method uses multi-scale soft threshold wavelet denoising technology, setting different denoising intensities for the scale and time period of the marked interference fluctuation points. In the specific implementation, a universal soft threshold function T(x) = sign(x) • max(|x|-λ, 0) is used, where the threshold λ is set based on the statistical energy at each scale. For example, the fourth scale is set to the mean plus 1.5 times the standard deviation. To avoid signal loss, the wavelet coefficients of non-interference points remain unchanged, and only specific areas are suppressed. The denoised coefficients are reconstructed using wavelet transforms to produce the interference-suppressed discharge signal fluctuation curve. Experimental results show that this method can effectively reduce the interference noise floor by over 10 dB while preserving the main discharge pulse, making the signal structure more prominent and clear. The resulting discharge signal curve exhibits a higher signal-to-noise ratio and enhanced temporal localization capability, providing a more reliable basis for subsequent compensation, restoration, and fault location. Due to shielding, reflection, and multipath effects during electromagnetic wave propagation within the GIS cavity, the signal at the actual discharge source is attenuated and distorted to varying degrees at the receiving end of each sensor. Therefore, signal backpropagation compensation is required based on the electromagnetic shielding assessment values ​​at each point.In the experiment, electromagnetic simulation was used to model the GIS cavity using CST or HFSS, simulating the propagation path and shielding effects of electromagnetic waves in various frequency bands. The simulation results yielded the equivalent transfer function H(f) from each sensor to the discharge source. Frequency-domain deconvolution of the interference-suppressed fluctuating signal was then performed to restore its original intensity. The compensation process needed to account for the nonlinear variation of the frequency response, so frequency-band compensation coefficients were employed. For example, an average compensation gain of +8 dB was set for the 300 MHz to 600 MHz band, while the gain was set to above +12 dB in high-frequency bands (>1 GHz). In practice, time reversal imaging (RTI) was used to compensate for the signal propagation path, taking into account the spatial distance between the sensor and the signal source, for more accurate source strength estimation. The resulting compensated discharge signal waveform approximated the original radiation characteristics of the discharge source, with more realistic and reliable parameters such as amplitude, rising edge, and width, providing critical data support for determining the partial discharge level and locating faults.

[0045] In this embodiment, reference Figure 4 The above is a flowchart of the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: The micro-vibration response parameters of GIS equipment during discharge excitation events are collected based on MEMS micro-seismic sensors. Calculating the vibration amplitude, transient intensity, and duration of the micro-vibration response parameters to obtain micro-vibration multimodal state characteristics; Perform time series correlation analysis on the original discharge signal based on the micro-vibration response parameters to identify the correlation between vibration and equipment discharge; According to the correlation law between vibration and equipment discharge, the structural vibration response time caused by discharge is calculated to obtain the micro-vibration response delay parameter; The actual influencing waveform points of the attenuation compensation discharge signal waveform are detected and marked according to the micro-vibration response delay parameters, and physical vibration interference is filtered based on the multi-modal state characteristics of micro-vibration to obtain a pure and optimized discharge signal waveform.

[0046] In this embodiment, because partial discharges exhibit transient high-frequency characteristics, often accompanied by electromagnetic pulses and weak mechanical vibrations, traditional electrical detection methods are susceptible to interference from complex electromagnetic environments. To address this, a MEMS accelerometer array with a response frequency band of 1 kHz to 20 kHz, a minimum detectable acceleration of 0.01 m / s², and a dynamic range of up to 90 dB was deployed at key points on the surface of the GIS equipment housing (such as near insulator mounting points and busbar connection points). The sampling frequency was set to 100 kHz to capture transient discharge excitations, and a synchronized clock system was used to ensure consistent timing of data from multiple sensors. The acquisition system, coupled with a high-impedance, low-noise preamplifier and an analog-to-digital conversion module, digitized the vibration signals in real time and stored them in an edge computing unit. Through offline analysis or real-time processing, initial micro-vibration response parameters, including vibration amplitude, waveform morphology, and frequency response, can be obtained, providing basic data support for subsequent signal feature calculation and pattern analysis. Multimodal features reflecting the partial discharge-induced vibration response are extracted from the microvibration data collected in the first step. These features primarily include amplitude, transient intensity, and duration. First, the vibration signal is bandpass filtered (1 kHz–20 kHz) to remove low-frequency interference from the environment and high-frequency interference from the system. Then, the signal is subjected to time-frequency analysis using short-time Fourier transform (STFT) and wavelet packet transform (WPT) to obtain the temporal variation of the frequency domain energy distribution. The vibration amplitude is obtained using the peak envelope detection method, with the peak value and root mean square (RMS) value recorded as amplitude features. For transient intensity, a Gabor filter bank is used to detect transient states in the signal, extracting energy mutation points and calculating their average slope and rate of change. For duration, a threshold method is used to identify the start and end points of the vibration signal. Event boundaries are determined based on points where the energy density is less than 10% of the maximum value, thereby calculating the effective duration of the microvibration under discharge excitation. Finally, the above feature sets are combined into a multi-dimensional state vector, forming the "micro-vibration multimodal state characteristics" of each discharge event, laying the foundation for distinguishing the difference between normal vibration and discharge-induced vibration.

[0047] A temporal correlation mechanism between microvibration signals and partial discharge events was established, revealing the causal relationship and statistical coupling characteristics between the two. High-frequency partial discharge signals (such as ultra-high frequency (UHF) and transient earth voltage (TEV)) were time-aligned with vibration signals acquired by MEMS. Using a synchronized clock system (with an accuracy better than 10 μs) as a benchmark, a sliding time window comparison strategy was employed to compare the onset of each discharge signal with the corresponding microvibration response time. In the experiment, a standard partial discharge source (such as an artificial tip-plate electrode defect) was installed within the GIS equipment. The experimental voltage was gradually increased between 20 and 80 kV to induce partial discharge events and simultaneously acquire electrical signals and vibration responses. Next, a cross-correlation function was used to calculate the maximum correlation value and delay position between the vibration and electrical signals to determine the degree of response matching. Statistical analysis (such as the Pearson coefficient and mutual information) was used to establish a quantitative relationship between vibration amplitude and partial discharge voltage amplitude, extract regular features, and form a multi-event sample-level "vibration-equipment discharge correlation law model," providing a theoretical basis for delay estimation and signal filtering. Based on the established correlation model, the mechanical response time of each partial discharge event, known as the "micro-vibration response delay parameter," is accurately estimated. Due to the complex mechanical wave propagation paths within the GIS equipment housing and internal structure, the discharge-induced mechanical response signal often exhibits delays of tens to hundreds of microseconds. This step utilizes a fusion of physical modeling and experimental data. Firstly, a finite element simulation model (such as ANSYS Modal & Harmonic Response) is constructed based on the GIS structure's material parameters (e.g., the speed of sound in the aluminum alloy housing is approximately 5000 m / s), geometry, and sensor placement to predict the time delay range from the discharge point to the sensor node. Secondly, combined with discharge-vibration correlation time difference data extracted from experimental samples, regression analysis and Bayesian estimation methods are used to correct the theoretical propagation time, resulting in a more realistic "response delay probability distribution model." Finally, the expected value is used as the "micro-vibration response delay parameter," and a specific delay time is assigned to each discharge event, guiding subsequent signal synchronization calibration and denoising. The original discharge signal is finely corrected and interference removed using the aforementioned delay parameters and multimodal state characteristics, ultimately extracting a pure, optimized discharge waveform. First, based on the "response delay parameter" obtained in step 4, the electrical signal is time-adjusted. This is compared to the expected vibration response time window for each partial discharge event to locate key points in the waveform that may be affected by vibration. Next, wavelet denoising (e.g., using Daubechies wavelets) is performed on the discharge signal waveform, and the vibration amplitude and duration are combined to determine whether it is a structural response or environmental interference.If a signal segment simultaneously satisfies the following conditions: ① the delayed position falls within the predetermined response time window; ② the vibration eigenvector matches the vibration-discharge state model; and ③ the vibration amplitude exceeds the background threshold, the waveform point is identified as a "physical vibration superposition interference point" and an adaptive filter is used to restore the amplitude and morphology. To avoid false filtering, segments with irrelevant vibration characteristics are retained, and envelope analysis and spectrum reconstruction techniques are used to restore the original discharge waveform. The resulting "optimized discharge signal waveform" has a higher signal-to-noise ratio and timing accuracy, enabling direct use in high-precision partial discharge location and pattern recognition, providing a reliable data foundation for GIS equipment condition monitoring and life assessment.

[0048] In this embodiment, step S4 includes the following steps: Multi-point signal intensity calculation is performed based on the pure optimized discharge signal waveform, and gradient fitting is performed to construct the internal discharge intensity gradient curve; Based on the internal discharge intensity gradient curve, the intensity attenuation direction is identified to obtain the discharge intensity attenuation direction; Analyze the partial discharge propagation path based on the discharge intensity attenuation direction to obtain the discharge propagation paths of multiple sampling points; The discharge propagation paths of multiple sampling points are input into a convolutional neural network structure and trained end-to-end to extract the spatial correlation patterns of different propagation paths; The discharge area is fuzzy located based on the spatial correlation pattern, and a probability distribution heat map of the discharge source area is constructed.

[0049] In this embodiment, after obtaining the purified discharge waveform, the goal of this step is to perform spatial intensity analysis on the signals acquired by multiple MEMS microseismic sensor measurement points installed on the surface of the GIS shell or at key locations within the structure, thereby reconstructing the propagation intensity distribution of the energy generated by the partial discharge on the structure. Specifically, no fewer than eight deployment points are selected to ensure signal field coverage and fitting accuracy. Each sampling point captures a discharge response waveform recorded with the same time base, and after noise removal and delay compensation, data synchronization is ensured. For each waveform, peak amplitude and effective mean square (RMS) are used as dual measurement parameters for the discharge response intensity, and its envelope area is also calculated as an energy indicator.

[0050] To perform gradient curve fitting, the actual spatial coordinates of the measurement points (calibrated using the GIS CAD model) and their corresponding signal intensities are combined to form a spatial intensity vector matrix. Multivariate high-order surface fitting (such as quadratic polynomial surfaces) or radial basis function interpolation (RBF) is used to model the three-dimensional signal intensity field and construct a "discharge intensity gradient curve." Experimental results show that for a typical GIS cylindrical structure (approximately 600 mm in diameter and 1.2 m in length), RBF fitting can stably reconstruct the energy attenuation path along the shell when the number of signal points is ≥10, providing a basis for subsequent direction identification. After completing the intensity gradient modeling, the current step aims to analyze the main attenuation direction of energy propagating outward from the source point. Because the mechanical energy induced by partial discharge propagates anisotropically in the GIS structure, the intensity gradient field often contains one or more main energy attenuation paths. First, the first-order partial derivative of the gradient curve is calculated to obtain the rate of change of intensity in each direction at each sampling point. Then, a gradient vector field is constructed, marking the local direction of the fastest attenuation at each location. Principal direction clustering (e.g., based on K-means or Mean Shift) is used to extract the set of dominant attenuation vectors, and a weighted average method is used to extract the "global attenuation principal direction" of the overall phenomenon. This direction generally points to the path connecting the discharge source and the response point and can be considered the primary channel for the outward radiation of vibration energy from the partial discharge source. Experimental measurements using an artificial discharge source (using a needle-plate electrode discharge source) show that the principal direction deviation obtained from different excitation locations can be controlled to within 5-15°, forming a clear directional ray field in the graphical interface, validating the guiding value of this direction identification in spatial positioning. After identifying the attenuation direction, the three-dimensional structure and material properties of the GIS are further combined to reversely trace the potential discharge propagation path for each measurement point signal. Given that the GIS structure consists of multiple metal shells, support insulators, lightning arresters, and other components, the propagation path may be deflected multiple times due to reflection, attenuation, and refraction. Therefore, path analysis is not based solely on straight-line directions but also needs to account for propagation differences between media. The attenuation direction of each sampling point is projected back onto the 3D CAD structural model. Based on common locations of discharge sources (such as within capacitor bushings, busbar branch openings, and fixed bracket ends), the shortest propagation path principle (based on a modified A algorithm or geometric ray tracing) is used to calculate the possible discharge source locations along the attenuation direction from each sampling point. Furthermore, to account for differences in structural materials, propagation velocity differences are set (e.g., 5000 m / s in metal areas and 2800 m / s in insulating materials), and propagation delay correction is incorporated. Ultimately, one or more propagation paths are established between each sensor point and the suspected source, forming a "discharge propagation path library" that provides the data foundation for subsequent deep learning processing. Deep learning methods are used to process the propagation path information between multiple sampling points, with the goal of automatically extracting the spatial characteristic patterns between the discharge source and each response point.Path information is converted into structured input data: each path forms a feature vector with its starting point, direction vector, propagation distance, signal strength attenuation rate, propagation time delay, etc. Multiple paths can be combined into a multi-channel input matrix, similar to the multi-channel image data input to a convolutional neural network (CNN).

[0051] A typical three-layer CNN architecture is employed: the first layer extracts geometric relationships between local path segments, the second layer extracts spatiotemporal synchronization features between multiple paths, and the third layer implements spatial fusion pattern recognition. During training, supervised training is performed using samples with known discharge source locations (e.g., multiple sampling of an artificial discharge source at a fixed location on a GIS test platform). The loss function uses a combination of mean squared error and KL divergence to ensure that the predicted spatial pattern matches the probability distribution of the true location. In experiments, a model built using the TensorFlow framework achieved an average localization accuracy of approximately 87% with 50 training samples and 15 validation samples, demonstrating the significant advantages of CNNs in mining spatial features between discharge propagation paths. After the CNN model is trained and the discharge propagation pattern is obtained, the purpose of this step is to apply the extracted spatial feature patterns to actual monitoring for fuzzy localization of unknown discharge sources. For each newly detected partial discharge event, the multi-point response signal path collected in real time is fed into the trained CNN model, which outputs a set of spatial distribution probability values ​​representing the probability of each candidate region being the true discharge source. These probability values ​​are mapped to corresponding locations in the GIS three-dimensional structural model, and spatial interpolation (e.g., three-dimensional Gaussian kernel density estimation) is performed to generate a continuous thermal distribution map. This heat map clearly reflects the most likely location of the discharge source, such as a high-probability cluster (>0.75) on the right side of the barrel near the support insulator. Furthermore, this heat map can be overlaid on the equipment visualization interface, assisting maintenance personnel in quickly inspecting, dismantling, or patrolling the target area, significantly improving the efficiency of PD location. Experimental verification shows that, in standard test GIS equipment, the error between the peak area of ​​the constructed heat map and the actual discharge source location is within 100 mm. This improves the reliability and resolution of spatial identification compared to traditional UHF positioning methods, making it highly suitable for intelligent detection and diagnosis of complex PD environments.

[0052] In this embodiment, the specific steps of step S5 are: Perform multi-point phase consistency optimization on the pure optimized discharge signal waveform to obtain a time-space phase consistent waveform; Perform multi-point synchronous RF positioning analysis on the time-space phase-consistent waveform to construct the three-dimensional time-series propagation trajectory of the discharge source; Based on the three-dimensional time series propagation trajectory of the discharge source, reverse discharge simulation is performed on the probability distribution heat map of the discharge source area to obtain the discharge simulation data of the initial discharge source; Based on the discharge simulation data of the initial discharge source, accurate multi-region discharge modeling is performed to construct an accurate discharge signal distribution trajectory.

[0053] In this embodiment, after obtaining pure, optimized discharge waveforms at multiple measurement points, the core task of this step is to perform phase consistency calibration to achieve subsequent time series trajectory restoration. This ensures that the discharge signals between the multiple measurement points meet the required synchronization accuracy in terms of time and phase. Although local discharge events are extremely instantaneous, during structural propagation, due to factors such as path length, dielectric differences, and signal reflections, phase offsets of microseconds or even greater can occur at different measurement points. Therefore, optimizing the phase consistency of the original signal is crucial to ensuring positioning accuracy and the credibility of the time series modeling.

[0054] First, a fast Fourier transform (FFT) is performed on the cleaned waveform at each measurement point to obtain frequency domain information, extracting the dominant frequency component and its phase angle. A reference signal (typically the measurement point with the highest signal-to-noise ratio) is used to establish a phase baseline, and phase-locking analysis is used to compare the phases of the same-frequency signals at other measurement points. Next, the instantaneous phase is obtained using a Hilbert transform, and a delay compensation algorithm is used to align all waveforms to a common phase starting point, ensuring that the transient characteristic peak at all measurement points is within ±1 μs of the same time calibration point. Based on the phase-calibrated multi-point signal waveforms, this step uses multi-sensor data similar to the TDOA (Time Difference of Arrival) multi-point synchronous positioning principle to construct the three-dimensional propagation path and time-series trajectory of the discharge signal within the GIS. Due to the complex GIS structure, uncertain discharge location, and non-uniform mechanical wave propagation velocity, direct positioning is subject to high-dimensional nonlinear errors. Therefore, this step combines phase difference information with a structural propagation model to achieve high-precision trajectory reconstruction. First, the propagation time of the main pulse is calculated by analyzing the arrival time of each measuring point (based on the maximum point or first transition point of the RMS envelope). Distance differences are estimated by combining the propagation velocities of different materials within the GIS (e.g., 5000 m / s for metal and 2700 m / s for insulation). Then, the time difference of orientation (TDOA) is used to construct bidirectional propagation path differences between the measuring points. These path differences are mapped to ellipsoidal isochronal surface constraints using the GIS CAD model. The intersection of all ellipsoidal surfaces is calculated in a three-dimensional structure, and the coordinates of the discharge starting point and its propagation path are determined using a minimum error point fitting method (e.g., Levenberg-Marquardt optimization). The arrival time sequence of each wavefront at each measuring point is recorded, forming a "three-dimensional temporal propagation trajectory of the discharge source." In a typical experimental scenario, using nine measuring points and a discharge in a 2.5 m × 1 m space, the maximum trajectory reconstruction error is kept within 3 cm, significantly exceeding the accuracy of traditional single-point positioning algorithms. This path is then subjected to inverse modeling and simulation, with the goal of using the propagation trajectory as the "result" to infer the "initial discharge source behavior model" that most likely generated this path. This step, similar to seismic inversion methods, converts observed data into source parameters, constructing a "simulated generation process" of the discharge source and enhancing understanding of the physical mechanisms of partial discharges. Based on the starting point location and time series in the propagation trajectory, an initial point source model is constructed. A finite element model is then built, combining material propagation properties (acoustic impedance and absorption coefficient) with GIS structural boundary conditions (such as wave reflection coefficient and geometric constraints). Inverse wavefield simulations are performed using the Acoustic / Structure Coupling Module in COMSOL or ANSYS, inputting propagation path information as constraints to determine the most likely initial signal waveform and source intensity.To enhance the match between simulation and real-world data, a particle swarm optimization (PSO) algorithm was introduced to automatically adjust simulation parameters (such as initial energy, angular distribution, and frequency content) to minimize the difference function between the simulated signal and the measured trajectory at each measurement point. This ultimately yielded a set of "discharge source simulation data," including the initial discharge waveform, source energy, and spatial radiation model. These data serve as key physical inputs for modeling the discharge propagation characteristics within the entire GIS. The discharge source data obtained through reverse simulation was used for forward propagation modeling and extended to multiple regions throughout the GIS space to establish a high-fidelity modeling system for the propagation of discharge events within complex structures. First, multiple possible discharge source points (such as busbar connections, bushing ends, and near grounding brackets) were set as simulation starting points. Using the obtained initial discharge source parameters, a three-dimensional simulation was performed within each region.

[0055] During the modeling process, a full-structure model was constructed using multi-physics coupled simulation software (such as ANSYS Multiphysics) to simulate the mechanical waves generated by electrical excitation under real-world operating conditions. Boundary loss models (acoustic absorption materials and structural damping) were incorporated to enhance propagation realism. A "discharge signal distribution trajectory" was output for each region, including the propagation path, energy attenuation spectrum, and time-delay distribution curve, and compared with the previously measured data. To improve the model's generalization, multiple historical data samples (≥30 sets of partial discharge data from different defect locations) were used for training and optimization, constructing a complete "precise discharge modeling database." In subsequent field monitoring, this database enables pattern matching between new discharge events and multiple regional models in the database, allowing for rapid identification of their source regions and propagation behavior. This precise modeling system provides core support for early fault warning, defect development trend assessment, and internal structural safety analysis in GIS equipment, enabling a transition from intelligent diagnosis at the "signal processing" level to "physical mechanism" level, truly enhancing the health management capabilities of GIS equipment throughout its lifecycle.

[0056] In this embodiment, the specific steps of step S6 are: Based on the precise discharge signal distribution trajectory, the discharge trajectory position of the metal shell electromagnetic reflection twin model is aligned and the discharge trajectory is mapped to construct a time-series discharge twin model of the GIS equipment. Based on the GIS equipment time-series discharge twin model, the discharge trend evolution of multiple parts is carried out to generate the discharge trend evolution characteristics of multiple parts; Insulation degradation status prediction is performed based on the discharge trend evolution characteristics of multiple parts to obtain the insulation degradation status characteristics of different parts; Based on the characteristics of insulation degradation, local abnormal events are simulated, local discharge risk assessment is performed, and a real-time discharge evolution warning image is constructed.

[0057] In this embodiment, the three-dimensional discharge distribution trajectory obtained from the precise modeling of the discharge signal is spatially mapped and dynamically registered to a digital GIS equipment structural model, thereby establishing an "electromagnetic twin model" with reflection characteristics and time-series mapping capabilities. Because the metal enclosure and multi-layered conductive components within the GIS exhibit complex reflection, refraction, and interference effects on both electromagnetic and structural waves, the twin model must incorporate not only the geometric structure but also the time-domain electromagnetic response characteristics. A detailed three-dimensional CAD model is constructed based on a real-world GIS structure (e.g., an ABB 252kV model). Based on this model, a physical twin environment is constructed, including key factors such as the inner wall metal reflection parameters (reflectivity set to 0.88-0.95), connection component boundary conditions, and electrical contact resistance modeling. Subsequently, an exact matching algorithm (such as ICP-Iterative Closest Point) is used to map the coordinates of the discharge signal propagation trajectory obtained earlier to the twin model, achieving consistent spatial registration. The finite-difference time-domain (FDTD) method is used to reconstruct the electromagnetic propagation paths in the twin model in the time domain, superimposing the propagation time series information of discharge events to form a "time-space coupled model." The resulting "GIS equipment time-series discharge twin model" simulates the dynamic response of the discharge signal propagating from the source to various locations at each moment, providing a realistic foundation for trend analysis and predictive reasoning of multi-source discharges. Continuous analysis of the evolutionary trends of different potential discharge source locations in the time dimension reveals the temporal evolution of their discharge characteristics. The evolutionary characteristics of multi-location discharge trends are mainly reflected in the following aspects: increasing frequency, increasing energy, expanding propagation paths, and varying signal delays. Monitoring points are first set at several key discharge areas selected within the twin model (such as the bushing-busbar connection, arrester joints, and the lower part of the supporting insulator), and real-time or historical discharge signal sequences are continuously input. Continuous discharge events are then analyzed using a sliding time window (e.g., a 1-minute window with a 30-second step size). Statistical characteristic parameters within each window are calculated, including the number of discharge pulses, average amplitude, location changes of the maximum energy point, and propagation time stability. Time series analysis methods (e.g., ARIMA models and time series convolutional networks (TCNs)) are used to model the temporal evolution of these characteristics, generating "trend evolution curves" for each region. Furthermore, a combination of K-means clustering and principal component analysis (PCA) is used to perform multidimensional dimensionality reduction on the evolutionary characteristics, identifying clustering patterns of discharge behavior at the trend level in different regions. Furthermore, by combining the physical properties of insulation materials with historical degradation samples, the degree of insulation degradation within the GIS is dynamically predicted, ultimately constructing an insulation "health map." Discharge characteristic mapping models corresponding to insulation degradation are established for different structural materials.For example, degradation in GIS sections using epoxy-resin-supported insulation often manifests as frequent, low-amplitude discharges early on, followed by a sudden energy surge. In contrast, areas using ceramic insulation may exhibit discharges with stable frequency but large amplitude fluctuations. Using a variety of typical insulation samples, a BP neural network was used to train a mapping between discharge trends and degradation levels, outputting a three-level health status rating (normal, warning, and severe). The evolutionary characteristics were fed into the prediction model as input vectors, and combined with the regional calibration function of the twin model, the predicted insulation degradation values ​​for each region were output. To improve prediction stability, a Markov chain was introduced to smooth the prediction results over time to prevent occasional outliers from interfering with trend analysis. Based on the local insulation degradation prediction results obtained above, the discharge risk evolution for future periods was simulated, ultimately generating a "discharge evolution warning image" within the GIS visualization platform, which serves as a key basis for maintenance decisions. Based on the predicted insulation degradation characteristics, the discharge response module within the twin model was invoked to perform Monte Carlo simulations of possible discharge events in the future. This simulation process uses factors such as the current degree of degradation, discharge intensity trends, structural propagation paths, and coupling effects between adjacent structures as inputs to randomly generate multiple samples of future discharge events. Using risk assessment models (e.g., Bayesian network-based risk propagation path analysis), the probability of a high-intensity partial discharge occurring within the next 24-72 hours is calculated for each region. This model comprehensively considers factors such as the discharge intensity accumulation factor, stress concentration in the electric field distribution, and the impact of structural temperature rise, and outputs a regional risk level distribution (low, medium, high, and very high). The risk assessment results are integrated with the GIS 3D structure to create a dynamically refreshable, color-coded "discharge evolution warning image." This image intuitively reflects the current equipment health status, risk concentration areas, and possible future local breakdown trends, supporting intelligent early warning from the remote operation and maintenance platform.

[0058] In this embodiment, a high-precision partial discharge detection system for a fully enclosed GIS device is provided, which is used to perform the high-precision partial discharge detection method for a fully enclosed GIS device as described above, including: An electromagnetic shielding evaluation module is used to perform three-dimensional structural modeling of the metal shell of the GIS equipment and perform electromagnetic shielding evaluation to obtain multi-point electromagnetic shielding evaluation values; An attenuation compensation module is used to collect the original discharge signal of each sampling point based on the electromagnetic wave sensor, and perform interference suppression processing and reverse propagation attenuation compensation according to the multi-point electromagnetic shielding evaluation value, so as to obtain an attenuation compensated discharge signal waveform; The physical vibration filtering module is used to collect micro-vibration response parameters based on the MEMS micro-seismic sensor, and to detect and mark the actual influencing waveform points and filter the physical vibration interference on the attenuation compensation discharge signal waveform, thereby obtaining a pure and optimized discharge signal waveform; The propagation path analysis module is used to analyze the partial discharge propagation path based on the pure optimized discharge signal waveform, perform fuzzy positioning of the discharge area, and construct a probability distribution heat map of the discharge source area; The reverse discharge simulation module is used to perform multi-point synchronous RF positioning analysis on the pure and optimized discharge signal waveform, and to perform reverse discharge simulation on the probability distribution heat map of the discharge source area to construct a precise discharge signal distribution trajectory; The degradation trend prediction module is used to analyze the discharge trend evolution of multiple locations based on the precise discharge signal distribution trajectory, predict the insulation degradation trend, and construct a real-time discharge evolution warning image.

[0059] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0060] The foregoing description is intended only to provide specific embodiments of the present invention, which are intended to enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest manner consistent with the principles and novel features disclosed herein.

Claims

1. A high-precision detection method for partial discharge of fully enclosed GIS equipment, characterized in that: The GIS equipment is configured to deploy multiple electromagnetic wave sensor sampling points inside and MEMS microseismic sensors on the surface; the steps include: Step S1: performing three-dimensional structural modeling of the metal shell of the GIS equipment and performing electromagnetic shielding evaluation to obtain multi-point electromagnetic shielding evaluation values; Step S2: collecting the original discharge signal of each sampling point based on the electromagnetic wave sensor, and performing interference suppression processing and reverse propagation attenuation compensation according to the multi-point electromagnetic shielding evaluation value, thereby obtaining an attenuation-compensated discharge signal waveform; Step S3: collecting micro-vibration response parameters based on the MEMS micro-seismic sensor, and performing actual influencing waveform point detection and marking on the attenuation compensation discharge signal waveform and filtering physical vibration interference, thereby obtaining a pure optimized discharge signal waveform; Step S4: Analyze the partial discharge propagation path based on the pure optimized discharge signal waveform, perform fuzzy positioning of the discharge area, and construct a probability distribution heat map of the discharge source area; Step S5: Perform multi-point synchronous radio frequency positioning analysis on the pure optimized discharge signal waveform, and perform reverse discharge simulation on the probability distribution heat map of the discharge source area to construct an accurate discharge signal distribution trajectory; Step S6: Evolve the discharge trend of multiple locations based on the precise discharge signal distribution trajectory, predict the insulation degradation situation, and construct a real-time discharge evolution warning image.

2. The high-precision detection method for partial discharge of fully enclosed GIS equipment according to claim 1 is characterized in that: The specific steps of step S1 are: Performing a three-dimensional structural analysis on the metal shell of the GIS device to obtain the three-dimensional structural characteristics of the surface of the GIS device; Conduct 3D modeling based on the 3D structural features of the GIS equipment surface and construct a metal shell model; Identifying the material properties of the metal shell, simulating the electromagnetic wave propagation and reflection distribution inside the shell for the metal shell model, and constructing an electromagnetic reflection twin model of the metal shell; Perform electromagnetic signal distribution coupling on the electromagnetic reflection twin model of the metal shell to obtain an electromagnetic signal distribution map; Electromagnetic shielding evaluation is performed based on the electromagnetic signal distribution diagram to obtain multi-point electromagnetic shielding evaluation values.

3. The high-precision detection method for partial discharge of a fully enclosed GIS device according to claim 1 is characterized in that: The specific steps of step S2 are: Based on multiple electromagnetic wave sensor sampling points, the original discharge signal of each sampling point is continuously collected; Performing a multi-scale wavelet transform on the original discharge signal to obtain time-frequency characteristics of the discharge signal at multiple scales; Performing time-frequency fluctuation analysis on each scale of the discharge signal time-frequency characteristics, and extracting the signal fluctuation curve of each scale; Identifying abnormal noise fluctuations on the signal fluctuation curve and marking interference source fluctuation points; Calculating the fluctuation intensity of the interference source fluctuation point, and performing interference suppression processing on the signal fluctuation curve to obtain an interference suppression discharge signal curve; The interference suppression discharge signal curve is subjected to reverse propagation attenuation compensation according to the multi-point electromagnetic shielding evaluation value, thereby obtaining an attenuation compensated discharge signal waveform.

4. The high-precision detection method for partial discharge of a fully enclosed GIS device according to claim 3 is characterized in that: The specific steps of performing reverse propagation attenuation compensation on the interference suppression discharge signal curve to obtain the attenuation compensation discharge signal waveform are: Define multiple timestamps; Extracting the original discharge signal of each collection point at each time node according to the multiple time stamps to obtain the original discharge signal at multiple moments; Calculating the signal differences at corresponding moments based on the original discharge signals at multiple moments to obtain signal difference values ​​at different moments; The signal difference values ​​at different times are averaged to obtain the average value of the signal difference; The signal propagation attenuation of the metal casing of the GIS equipment is analyzed based on the average signal difference and the multi-point electromagnetic shielding evaluation value, thereby obtaining the propagation attenuation characteristics of the metal casing signal; Based on the propagation attenuation characteristic, reverse propagation attenuation compensation is performed on the interference suppression discharge signal curve, thereby obtaining an attenuation compensated discharge signal waveform.

5. The high-precision detection method for partial discharge of fully enclosed GIS equipment according to claim 1 is characterized in that: The specific steps of step S3 are: The micro-vibration response parameters of GIS equipment during discharge excitation events are collected based on MEMS micro-seismic sensors. Calculating the vibration amplitude, transient intensity, and duration of the micro-vibration response parameters to obtain micro-vibration multimodal state characteristics; Perform time series correlation analysis on the original discharge signal based on the micro-vibration response parameters to identify the correlation between vibration and equipment discharge; According to the correlation law between vibration and equipment discharge, the structural vibration response time caused by discharge is calculated to obtain the micro-vibration response delay parameter; The actual influencing waveform points of the attenuation compensation discharge signal waveform are detected and marked according to the micro-vibration response delay parameters, and physical vibration interference is filtered based on the multi-modal state characteristics of micro-vibration to obtain a pure and optimized discharge signal waveform.

6. The high-precision detection method for partial discharge of fully enclosed GIS equipment according to claim 1 is characterized in that: The specific steps of step S4 are: Multi-point signal intensity calculation is performed based on the pure optimized discharge signal waveform, and gradient fitting is performed to construct the internal discharge intensity gradient curve; Based on the internal discharge intensity gradient curve, the intensity attenuation direction is identified to obtain the discharge intensity attenuation direction; Analyze the partial discharge propagation path based on the discharge intensity attenuation direction to obtain the discharge propagation paths of multiple sampling points; The discharge propagation paths of multiple sampling points are input into a convolutional neural network structure and trained end-to-end to extract the spatial correlation patterns of different propagation paths; The discharge area is fuzzy located based on the spatial correlation pattern, and a probability distribution heat map of the discharge source area is constructed.

7. The high-precision detection method for partial discharge of fully enclosed GIS equipment according to claim 1 is characterized in that: The specific steps of step S5 are: Perform multi-point phase consistency optimization on the pure optimized discharge signal waveform to obtain a time-space phase consistent waveform; Perform multi-point synchronous RF positioning analysis on the time-space phase-consistent waveform to construct the three-dimensional time-series propagation trajectory of the discharge source; Based on the three-dimensional time series propagation trajectory of the discharge source, reverse discharge simulation is performed on the probability distribution heat map of the discharge source area to obtain the discharge simulation data of the initial discharge source; Based on the discharge simulation data of the initial discharge source, accurate multi-region discharge modeling is performed to construct an accurate discharge signal distribution trajectory.

8. The high-precision detection method for partial discharge of fully enclosed GIS equipment according to claim 1 is characterized in that: The specific steps of step S6 are: Based on the precise discharge signal distribution trajectory, the discharge trajectory position of the metal shell electromagnetic reflection twin model is aligned and the discharge trajectory is mapped to construct a time-series discharge twin model of the GIS equipment. Based on the GIS equipment time-series discharge twin model, the discharge trend evolution of multiple parts is carried out to generate the discharge trend evolution characteristics of multiple parts; Insulation degradation status prediction is performed based on the discharge trend evolution characteristics of multiple parts to obtain the insulation degradation status characteristics of different parts; Based on the characteristics of insulation degradation, local abnormal events are simulated, local discharge risk assessment is performed, and a real-time discharge evolution warning image is constructed.

9. A high-precision partial discharge detection system for fully enclosed GIS equipment, characterized in that: The method for high-precision detection of partial discharge of a fully enclosed GIS device according to claim 1 comprises: An electromagnetic shielding evaluation module is used to perform three-dimensional structural modeling of the metal shell of the GIS equipment and perform electromagnetic shielding evaluation to obtain multi-point electromagnetic shielding evaluation values; An attenuation compensation module is used to collect the original discharge signal of each sampling point based on the electromagnetic wave sensor, and perform interference suppression processing and reverse propagation attenuation compensation according to the multi-point electromagnetic shielding evaluation value, so as to obtain an attenuation compensated discharge signal waveform; The physical vibration filtering module is used to collect micro-vibration response parameters based on the MEMS micro-seismic sensor, and to detect and mark the actual influencing waveform points and filter the physical vibration interference on the attenuation compensation discharge signal waveform, thereby obtaining a pure and optimized discharge signal waveform; The propagation path analysis module is used to analyze the partial discharge propagation path based on the pure optimized discharge signal waveform, perform fuzzy positioning of the discharge area, and construct a probability distribution heat map of the discharge source area; The reverse discharge simulation module is used to perform multi-point synchronous RF positioning analysis on the pure and optimized discharge signal waveform, and to perform reverse discharge simulation on the probability distribution heat map of the discharge source area to construct a precise discharge signal distribution trajectory; The degradation trend prediction module is used to analyze the discharge trend evolution of multiple locations based on the precise discharge signal distribution trajectory, predict the insulation degradation trend, and construct a real-time discharge evolution warning image.

Images