Partial discharge detection method and device for totally-enclosed vacuum environment-friendly GIS equipment

Abstract

The invention relates to the technical field of GIS equipment, and discloses a partial discharge detection method and device for totally-enclosed vacuum environment-friendly GIS equipment, and the method comprises the steps: constructing a dual detection circuit system of the GIS equipment, and building an equipment impedance characteristic model which comprises housing distribution inductance and internal conductor distribution inductance; acquiring a first multi-mode discharge signal in a vacuum insulation environment, performing empirical mode decomposition to obtain a plurality of eigenmode function components, and synchronously recording environment characteristic parameters; constructing a discharge signal identification model; and identifying the second multi-mode discharge signal of the GIS equipment through the discharge signal identification model, and determining the internal defect three-dimensional coordinate and the hazard level of the GIS equipment, so that intelligent identification of discharge characteristics is realized, the calculation efficiency is improved, the anti-interference capability is enhanced, the three-dimensional coordinate of the internal defect of the GIS equipment can be accurately calculated, and the reliability of the GIS equipment is improved. And comprehensive evaluation of defect types and hazard levels is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of GIS equipment, and in particular to a partial discharge detection method and device for fully enclosed vacuum environmentally friendly GIS equipment. Background Art

[0002] In recent years, fully enclosed vacuum environmentally friendly GIS equipment has been widely used in power systems due to its environmental performance and compact size, but its internal partial discharge problem is still a key factor affecting the safe and stable operation of the equipment. Traditional partial discharge detection methods mainly use single-band detection technology, such as high-frequency detection or UHF sensor detection in the IEC 60270 standard. This single detection method cannot take into account the needs of discharge quantitative measurement and precise positioning at the same time, especially in complex vacuum insulation environments. The effect is even more limited. With the large-scale application of fully enclosed vacuum environmentally friendly GIS equipment, the multi-mode discharge phenomenon caused by the multi-point grounding structure has become increasingly prominent. This complex discharge mode makes the traditional discharge signal processing algorithm have technical bottlenecks such as low computational efficiency, insufficient environmental adaptability and weak anti-interference ability when dealing with large-scale detection problems with time-varying characteristics and environmental interference.

[0003] In fully enclosed vacuum environmentally friendly GIS equipment, broadband oscillation has become an important problem in signal analysis in partial discharge detection. Low-frequency discharge signals and high-frequency discharge signals have been widely reported around the world, but in more complex multi-point grounding systems, multiple independent discharge signal components may appear at the same monitoring point at the same time, forming a multi-mode discharge phenomenon. The research on this complex discharge problem is still very scarce. Traditional discharge feature recognition methods mainly rely on empirical rules and fixed threshold judgments. They have poor adaptability to the dynamic changes of signal features in a vacuum environment, and it is difficult to effectively deal with the interference and superposition of different types of discharge sources, resulting in insufficient reliability of detection results. Summary of the invention

[0004] The present invention provides a partial discharge detection method and device for a fully enclosed vacuum environmentally friendly GIS device. The present invention realizes intelligent recognition of discharge characteristics, improves calculation efficiency and enhances anti-interference ability, can accurately calculate the three-dimensional coordinates of internal defects of GIS equipment, and realizes comprehensive evaluation of defect types and hazard levels.

[0005] In a first aspect, the present invention provides a method for detecting partial discharge of a fully enclosed vacuum environmentally friendly GIS device, the method comprising: Construct a dual detection circuit system for GIS equipment and establish an equipment impedance characteristic model including the distributed inductance of the shell and the distributed inductance of the internal conductor; Based on the device impedance characteristic model, the first multi-mode discharge signal of the GIS device in a vacuum insulation environment is collected and empirical mode decomposition is performed to obtain multiple intrinsic mode function components, and environmental characteristic parameters are simultaneously recorded; Constructing a discharge signal recognition model based on the multiple eigenmode function components and the environmental characteristic parameters; The second multi-mode discharge signal of the GIS device is identified by the discharge signal identification model to determine the three-dimensional coordinates and hazard level of the internal defects of the GIS device.

[0006] In a second aspect, the present invention provides a partial discharge detection device for a fully enclosed vacuum environmentally friendly GIS device, the partial discharge detection device for a fully enclosed vacuum environmentally friendly GIS device comprising: The detection module is used to construct a dual detection circuit system for GIS equipment and establish an equipment impedance characteristic model including the distributed inductance of the shell and the distributed inductance of the internal conductor; A decomposition module is used to collect the first multi-mode discharge signal of the GIS device in a vacuum insulation environment based on the device impedance characteristic model and perform empirical mode decomposition to obtain multiple intrinsic mode function components and simultaneously record environmental characteristic parameters; A construction module, used for constructing a discharge signal recognition model based on the multiple eigenmode function components and the environmental characteristic parameters; The identification module is used to identify the second multi-mode discharge signal of the GIS device through the discharge signal identification model, and determine the three-dimensional coordinates and hazard level of the internal defects of the GIS device.

[0007] In the technical solution provided by the present invention, the dual detection circuit system combining high-frequency detection and UHF detection constructed by the present invention realizes full-spectrum monitoring of discharge signals, which can not only quantitatively measure the amount of discharge charge, but also accurately locate the discharge source, effectively solving the problem of incomplete information of a single detection method. Secondly, by injecting a sweep frequency signal into the GIS device and collecting the response signal, an equipment impedance characteristic model including the shell distributed inductance and the internal conductor distributed inductance is established, accurately characterizing the electromagnetic wave propagation characteristics inside the device. The discharge signal is processed by the empirical mode decomposition technology to effectively separate multiple discharge types existing at the same time, overcoming the limitations of traditional single mode recognition. The method synchronously records the environmental characteristic parameters and associates them with the discharge signal, establishes a "discharge signal-environmental parameter" mapping relationship, and enables the detection system to adaptively adjust the recognition strategy according to environmental changes. The constructed PTB model integrates a variable-length sequence feature extraction module, a Transformer encoder network, and a Bayesian network reasoning module, and is trained through a proximal strategy optimization algorithm to realize intelligent recognition of discharge features, improve computational efficiency, and enhance anti-interference ability. Based on the UHF sensor array and hyperbolic positioning algorithm, combined with the PTB model recognition results, the three-dimensional coordinates of internal defects of the equipment can be accurately calculated, realizing a comprehensive assessment of defect type and hazard level. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0009] Figure 1 A schematic diagram of the steps of a partial discharge detection method for a fully enclosed vacuum environmentally friendly GIS device according to an embodiment of the present invention; Figure 2 It is a structural schematic diagram of a partial discharge detection device for a fully enclosed vacuum environmentally friendly GIS device in an embodiment of the present invention. DETAILED DESCRIPTION

[0010] The embodiment of the present invention provides a partial discharge detection method and device for a fully enclosed vacuum environmentally friendly GIS equipment. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0011] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 An embodiment of a partial discharge detection method for a fully enclosed vacuum environmentally friendly GIS device in an embodiment of the present invention includes: Step S1, constructing a dual detection circuit system of GIS equipment, and establishing an equipment impedance characteristic model including shell distributed inductance and internal conductor distributed inductance; It is understandable that the execution subject of the present invention can be a partial discharge detection device of a fully enclosed vacuum environmentally friendly GIS device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0012] Specifically, the high-frequency detection part is formed by integrating different hardware modules, combining standard measurement impedance, high-frequency coupling capacitor, digital bandpass filter, high-frequency signal amplifier and high-speed data acquisition card. The standard measurement impedance provides electrical characteristic test data and measures the impedance characteristics of the equipment. The high-frequency coupling capacitor helps to transmit the high-frequency signal to various components of the GIS equipment, while the digital bandpass filter effectively filters out unnecessary frequency signals and focuses on the key signal frequency band of the discharge signal. The high-frequency signal amplifier appropriately amplifies the signal to ensure that the signal strength is high enough for further processing. The high-speed data acquisition card is used to quickly collect these signals, ensure the high time resolution of signal capture, and adapt to the rapid change characteristics of the discharge signal in the GIS equipment. On the basis of the high-frequency detection part, at least three disc-shaped UHF sensors are selected to detect UHF signals. The disc-shaped UHF sensor sensitively captures the signal of partial discharge in the high-frequency range, and its superior sensitivity and directionality enable it to accurately identify the electromagnetic waves generated by partial discharge. These signals are transmitted to the UHF bandpass filter through the UHF signal coupling system, and the bandpass filter filters out unnecessary frequency band signals and retains only the key signals of partial discharge. The UHF low-noise amplifier performs necessary amplification on the signal to ensure that the signal quality is high enough for accurate analysis. The UHF signal digitization module converts the analog signal into a digital signal. A synchronous trigger unit is selected to coordinate the signal capture and processing between the high-frequency detection part and the UHF detection part. The synchronous trigger mechanism ensures the timing consistency between different signal acquisition modules and prevents data deviation caused by the delay difference of signal capture. The first detection circuit system is subjected to impedance matching to avoid reflection loss during signal transmission and ensure that the signal is efficiently and stably transmitted to different parts of the equipment. The second detection circuit system installs the disc-type UHF sensor on the insulator flange of the GIS equipment. As the interface between the equipment and the external environment, the insulator flange is a key area where partial discharge is prone to occur. The installed UHF sensor is connected to the UHF detection part through a shielded cable to form the second detection circuit system. On this basis, the second detection circuit system is subjected to electromagnetic shielding to prevent external electromagnetic interference from affecting signal acquisition. The design of the electromagnetic shielding layer adopts a grounded metal shielding layer, which is connected to the earth through multi-point grounding, effectively eliminating external electromagnetic noise and improving the purity and accuracy of the signal. The dual detection circuit system forms an efficient and stable signal capture and processing network, which monitors the partial discharge of the equipment in real time and reflects the electrical characteristics inside the GIS equipment. On this basis, by injecting a sweep frequency signal into the GIS equipment to stimulate the electromagnetic response inside the equipment and collect equipment impedance characteristic data, a device impedance characteristic model including the shell distributed inductance and the internal conductor distributed inductance is constructed to describe the electromagnetic environment inside the equipment.

[0013] The frequency sweep is injected into the GIS device through the test port of the device to perform frequency sweep. The function of the sweep signal is to stimulate various parts of the device, especially the electromagnetic field response inside the device, so as to effectively extract the frequency response characteristics of the device. The injected sweep signal covers a wide frequency range, ensuring that the electromagnetic response of different frequency bands is captured, and then the electromagnetic characteristics of the device are comprehensively analyzed. The frequency sweep process is realized through the test port and completed by a dedicated signal source and frequency generator. The sweep signal generated by the signal source is transmitted to the inside of the GIS device through a cable to stimulate the electrical system inside the device. After the signal is injected, the dual detection circuit system synchronously collects the response signal of the GIS device. The two parts of the dual detection circuit system - the high-frequency detection part and the UHF detection part - capture the signals of the high-frequency and UHF frequency bands respectively. In order to ensure the accuracy and integrity of the signal, the collected data of these two parts are time synchronized, and the high-frequency response signal is accurately compared and analyzed with the UHF response signal. The synchronously collected signals form the target response signal data set. The target response signal data set is analyzed in the frequency domain. The time domain signal is converted into the frequency domain signal using the fast Fourier transform technology to obtain the frequency response characteristics of the GIS equipment, revealing the electrical behavior characteristics of the equipment at different frequencies, especially the amplitude and phase changes of the signal. Through the frequency response characteristics, various electrical parameters inside the GIS equipment are extracted, especially the distributed parameters related to the electromagnetic characteristics. These parameters include the distributed inductance of the shell, the distributed inductance of the internal conductor, the distributed capacitance between the internal conductor and the shell, the grounding resistance of the shell grounding point, the equivalent capacitance of the insulator, and the equivalent parallel resistance of the insulator. By analyzing these parameters, the electromagnetic field distribution characteristics of the equipment are described, providing data support for the impedance characteristic modeling of the equipment. The distributed parameter set is substituted into the Maxwell equations. The Maxwell equations are the basic equations that describe the behavior of the electromagnetic field and can simulate the propagation characteristics of electromagnetic waves in space. By substituting the previously extracted distributed parameters into the equations, the propagation time delay, attenuation characteristics, and reflection coefficient of the electromagnetic waves inside the equipment are calculated. The propagation time delay reflects the time required for the electromagnetic wave to propagate inside the equipment, the attenuation characteristics reveal the attenuation of the signal strength during the propagation process, and the reflection coefficient indicates the degree of reflection of the signal on the electrical interfaces inside the equipment. By parameterizing these electromagnetic propagation characteristic data, an equipment impedance characteristic model describing the electromagnetic characteristics of GIS equipment is established.

[0014] Step S2: based on the device impedance characteristic model, collect the first multi-mode discharge signal of the GIS device in the vacuum insulation environment and perform empirical mode decomposition to obtain multiple intrinsic mode function components, and simultaneously record the environmental characteristic parameters; Specifically, the rated operating voltage is applied to the GIS equipment to put the equipment in an actual working state, reflecting the partial discharge signal generated during its operation. When the rated operating voltage is applied, the dual detection circuit system is used to continuously monitor the GIS equipment to obtain the original discharge signal. The high-frequency detection part and the UHF detection part of the dual detection circuit system capture the partial discharge signals of different frequency bands respectively. The amplitude of the original discharge signal is calibrated based on the equipment impedance characteristic model. Due to the complex electromagnetic environment inside the GIS equipment, the partial discharge signal will be distorted during the propagation process due to factors such as medium characteristics, electromagnetic interference, reflection and attenuation. The amplitude of the original signal is corrected using the equipment impedance characteristic model to eliminate the measurement error caused by changes in sensor sensitivity or electromagnetic environment. According to the electromagnetic propagation characteristic data of the GIS equipment, the attenuation and distortion generated during the signal propagation process are compensated so that the signal truly reflects the original characteristics of the partial discharge without being affected by the loss or distortion on the propagation path. After amplitude calibration and distortion compensation, the discharge signal is obtained. The calibrated discharge signal is preliminarily screened to extract the effective discharge signal and eliminate irrelevant signals. Since GIS equipment is affected by external electromagnetic interference or noise during operation, the directly collected signals contain interference signals from non-local discharge sources such as environmental noise, switching operation, arc discharge, etc. Through filtering, statistical analysis and pattern recognition, the signal generated by the internal discharge of GIS equipment is screened out as the first multi-mode discharge signal. The screened first multi-mode discharge signal is subjected to empirical mode decomposition to obtain multiple intrinsic mode function components. Empirical mode decomposition is an adaptive signal decomposition method that decomposes complex nonlinear and non-stationary signals into a series of intrinsic mode functions (IMFs) with different frequency characteristics, so that discharge signals of different modes can be effectively distinguished. Through empirical mode decomposition, different types of discharge modes (such as creepage discharge, suspension discharge, internal partial discharge, etc.) are identified through feature extraction and classification algorithms, thereby improving the diagnostic capability of the detection system. At the same time, the analysis of the intrinsic mode function components is used to extract the key features of the discharge signal, such as instantaneous frequency, energy distribution, etc., in order to evaluate the operating status and potential failure risks of GIS equipment. While analyzing the signal, the environmental characteristic parameters of the GIS equipment are synchronously collected to comprehensively evaluate the influencing factors of the discharge signal and improve the accuracy of signal recognition. The collection of environmental characteristic parameters involves multiple key variables, including ambient temperature, ambient humidity, equipment temperature, equipment internal pressure, ambient electromagnetic field strength, and equipment vibration parameters, etc. These parameters directly affect the insulation state, partial discharge characteristics, and discharge signal propagation characteristics of GIS equipment.

[0015] The first multi-mode discharge signal is preprocessed, and signal processing techniques such as filtering, denoising, and normalization are used to remove high-frequency interference and low-frequency trend drift, so that the discharge signal is more stable and reflects the local discharge characteristics of the GIS equipment. The empirical mode decomposition algorithm is applied to the preprocessed discharge signal, and the signal is iteratively screened to extract local extreme points and construct upper and lower envelopes. Empirical mode decomposition is a method of decomposing complex non-stationary signals into a series of intrinsic mode function components (IMF) by adaptive decomposition. In each iteration, all local extreme points are found from the preprocessed discharge signal, and then the upper and lower envelopes of the signal are constructed by cubic spline interpolation, and the mean curve of the envelope is calculated. The mean curve is subtracted from the original signal to obtain a preliminary decomposition result. This process is repeated until the extracted components meet the IMF condition, that is, the local extreme points of the signal are evenly distributed and the envelope mean curve approaches zero. A termination judgment is performed on the preliminary decomposition result to determine whether the decomposition condition is met. If the currently decomposed signal still has a significant trend component, the iterative decomposition continues until all the intrinsic mode function components meet the definition of IMF. The output results of the empirical mode decomposition include M initial intrinsic mode function components (IMF) and a residual term, where the IMF components represent the different modal characteristics of the discharge signal, and the residual term reflects the long-term trend component of the signal. Hilbert transform is performed on each initial intrinsic mode function component to extract its characteristic parameters such as instantaneous frequency, spectral entropy, energy ratio and signal complexity. The Hilbert transform provides the instantaneous spectrum information of the signal, so that the frequency characteristics of different IMF components can be clearly distinguished. The instantaneous frequency can describe the time-varying characteristics of the signal, the spectral entropy is used to measure the degree of chaos of the signal spectrum, the energy ratio characterizes the energy proportion of different IMF components, and the signal complexity is used to evaluate the mode complexity of the discharge signal. These characteristic parameter sets are used to further analyze the propagation mode of the discharge signal to distinguish different types of discharge phenomena, such as surface discharge, suspension discharge, internal insulation discharge, etc. Based on the equipment impedance characteristic model, the propagation mode matching degree of each IMF component is calculated to determine their propagation characteristics in the GIS equipment. Since the electromagnetic environment inside the GIS equipment is affected by the structural morphology, impedance characteristics and conductor layout, different types of discharge signals will propagate in different ways. The propagation mode matching calculation is used to determine whether each IMF component conforms to the known electromagnetic propagation mode of the GIS equipment. The propagation mode matching calculation is based on the group velocity, phase velocity, reflection coefficient and attenuation characteristics of the signal. The propagation parameters of each IMF component are compared with the equipment impedance characteristic model to determine the similarity of its propagation mode. According to the propagation mode matching results, the initial intrinsic mode function components are grouped and reconstructed to obtain multiple separated intrinsic mode function components.Different types of discharge signals have large differences in propagation modes. Grouping them according to the degree of matching can effectively separate low-frequency background noise, high-frequency interference and real discharge signals, making partial discharge detection of GIS equipment more accurate and reliable.

[0016] Step S3, constructing a discharge signal recognition model based on multiple eigenmode function components and environmental characteristic parameters; Specifically, multiple intrinsic mode function components and environmental characteristic parameters are feature separated to form a time domain feature set and a frequency domain feature set. Since the local discharge signal has time-varying and non-stationary characteristics, the time domain features mainly include statistical features such as instantaneous amplitude, mean, standard deviation, skewness, and kurtosis, while the frequency domain features extract the signal's power spectrum density, center frequency, bandwidth, spectral entropy and other information through Fourier transform, Hilbert transform and other methods. Environmental characteristic parameters such as temperature, humidity, pressure, electromagnetic interference, etc. also need to be included in the feature set to supplement the ability to interpret signal characteristics and improve the adaptability of the model. A variable-length sequence feature extraction module is constructed based on the time domain feature set, which performs deep learning modeling on the time series characteristics of the discharge signal. Since the duration and waveform shape of the local discharge signal vary with environmental conditions, variable-length sequence modeling is used to better adapt to discharge modes of different time scales. The module includes a multi-scale convolution layer, an attention pooling layer, and a greedy sequence selector. The multi-scale convolution layer captures the local features of the signal in different time windows and effectively extracts short-term and long-term dependencies through a multi-scale receptive field. The attention pooling layer highlights the features of key time segments through an adaptive weighting method to improve the model's sensitivity to local discharge patterns. The greedy sequence selector is used to screen the most representative time series segments to reduce computational complexity and remove redundant information. Based on the output of the time domain feature set and the variable-length sequence feature extraction module, a Transformer encoder network is constructed to fully mine the global features of the discharge signal. The Transformer encoder network adopts the structure of a position encoding layer, a multi-head self-attention layer, and a feedforward neural network. The position encoding layer is used to retain time series information so that the model can understand the temporal dependencies of the signal. The multi-head self-attention layer calculates the correlation between signal segments, so that the model can effectively capture long-term dependency features and avoid the gradient vanishing problem of traditional recurrent neural networks in long sequence processing. The feedforward neural network is used for nonlinear feature transformation to improve feature expression capabilities, thereby enhancing the ability to recognize complex local discharge patterns. Based on the output features of the Transformer encoder network, a Bayesian network inference module is constructed. This module uses the advantages of the probabilistic graphical model to realize the probabilistic reasoning and classification of the discharge signal pattern. The Bayesian network inference module includes a conditional probability table, a probability propagation algorithm, and a posterior probability calculation unit. The conditional probability table is used to store the joint probability distribution between the discharge pattern and the environmental parameters; the probability propagation algorithm derives the most likely discharge category by calculating the marginal probability and the maximum posterior probability; the posterior probability calculation unit is used to dynamically update the confidence of the model, so that the model can be adaptively adjusted when facing new data, thereby improving the robustness and accuracy of reasoning. In order to optimize the discharge signal recognition model and improve the generalization ability of the model, the output results of the Bayesian network inference module are used as reward signals to construct a proximal policy optimization training system.The proximal policy optimization training system continuously optimizes the decision-making ability of the model through reinforcement learning. It consists of a policy network, a value network and an optimizer. The policy network is used to generate classification decisions for discharge signals; the value network is used to evaluate the pros and cons of the current decision and provide optimization directions; the optimizer adjusts the network parameters based on the gradient descent method to maximize the cumulative reward signal. Since the proximal policy optimization method uses truncated importance sampling to limit the amplitude of policy updates, it is more stable than the traditional reinforcement learning algorithm and can effectively avoid the problem of the policy falling into the local optimum. After the training is completed, the output of the PPO training system is verified and the model is selected using K-fold cross validation. By dividing the data set into K subsets, one of them is selected as the test set in turn, and the rest are selected as the training set, and the training and testing are repeated K times, the performance of the model under different data distributions is effectively evaluated, and the risk of overfitting is reduced. The model with the best performance is selected through K-fold cross validation, and the discharge signal recognition model for partial discharge detection of GIS equipment is obtained.

[0017] Step S4: Identify the second multi-mode discharge signal of the GIS device through the discharge signal recognition model to determine the three-dimensional coordinates and hazard level of the internal defects of the GIS device.

[0018] Specifically, the GIS equipment is monitored online. The second multi-mode discharge signal is collected through a dual detection circuit system, and preprocessed and empirical mode decomposition are performed. The preprocessing process includes steps such as noise removal, filtering, and normalization to ensure that the signal quality is high enough for subsequent analysis. The empirical mode decomposition algorithm is applied to decompose the signal to obtain the intrinsic mode function components for testing. These intrinsic mode function components contain the time-frequency characteristics of the discharge signal. The test intrinsic mode function components are input into the discharge signal recognition model, and the time-frequency characteristics of the signal are extracted through the variable-length sequence feature extraction module. The variable-length sequence feature extraction module uses a multi-scale convolution layer to capture the changes of the signal at different time scales, and weights the key features through the attention mechanism to enhance the effective information in the discharge signal. In this way, the model accurately extracts the core information related to the discharge characteristics in the complex signal. After being processed by the multi-head self-attention layer, the model enhances the long-term dependency of the signal and extracts richer time-frequency features to classify the signal more accurately. Among them, the Bayesian network inference module obtains the discharge type recognition result by calculating the probability distribution of each discharge type. The result is based on Bayesian theory, obtained through conditional probability table and probability propagation algorithm, and the relationship between different features is comprehensively considered to provide high confidence prediction for the type classification of discharge signals. Through this process, it is identified whether there are different types of discharge phenomena such as metal suspended particles, insulator surface discharge, conductor tip discharge, insulator internal air gap discharge or poor contact discharge in the GIS equipment. Through at least three disc-type UHF sensors in the UHF detection part, the discharge signal received by each sensor is collected, and the time when the signal arrives at each sensor is recorded. By comparing the time difference of the signals received by these sensors, the time difference matrix is ​​calculated to reflect the relative time delay of the discharge signal propagating between the sensors, providing necessary data support for defect location. Through the calculation of the time difference matrix, the location of the discharge source is deduced, and the three-dimensional coordinates of the internal defects of the equipment are determined. The time difference matrix is ​​substituted into the hyperbolic positioning equation group. The hyperbolic positioning method calculates the specific location of the discharge source in space based on the time difference of the signals received by multiple sensors. Since the equation group is nonlinear, it is linearized by Taylor series expansion method to facilitate solution. Through the iterative process of weighted least squares method, the three-dimensional coordinates of the defect are effectively obtained. Determine the type of defect based on the discharge type identification results. Different defect types, such as metal suspended particles, insulator surface discharge, conductor tip discharge, etc., have different electromagnetic characteristics and discharge behaviors, so the defects are classified according to the identification results. Each defect type corresponds to a different hazard level. In order to evaluate the hazard of the defect, the discharge repetition rate, discharge phase stability, discharge amplitude trend, and discharge energy density are calculated based on the three-dimensional coordinates and type of the defect.The discharge repetition rate reflects the frequency of defect occurrence, the discharge phase stability describes the phase change of the signal during the discharge process, the discharge amplitude trend reveals the change of discharge intensity, and the discharge energy density reflects the accumulation of discharge energy. These parameters comprehensively reflect the severity and potential risks of the discharge phenomenon. By weighted fusion of these parameters, the hazard level of the discharge defect is obtained.

[0019] In the embodiment of the present invention, the dual detection circuit system combining high-frequency detection and UHF detection constructed by the present invention realizes full-spectrum monitoring of discharge signals, which can quantitatively measure the discharge charge and accurately locate the discharge source, effectively solving the problem of incomplete information of a single detection method. Secondly, by injecting a sweep frequency signal into the GIS device and collecting the response signal, an equipment impedance characteristic model including the shell distributed inductance and the internal conductor distributed inductance is established, accurately characterizing the electromagnetic wave propagation characteristics inside the device. The discharge signal is processed by the empirical mode decomposition technology to effectively separate multiple discharge types existing at the same time, overcoming the limitations of traditional single mode recognition. The method synchronously records the environmental characteristic parameters and associates them with the discharge signal, establishes a "discharge signal-environmental parameter" mapping relationship, and enables the detection system to adaptively adjust the recognition strategy according to environmental changes. The constructed PTB model integrates a variable-length sequence feature extraction module, a Transformer encoder network, and a Bayesian network reasoning module, and is trained through a proximal strategy optimization algorithm to realize intelligent recognition of discharge features, improve computational efficiency, and enhance anti-interference ability. Based on the UHF sensor array and hyperbolic positioning algorithm, combined with the PTB model recognition results, the three-dimensional coordinates of internal defects of the equipment can be accurately calculated, realizing a comprehensive assessment of defect type and hazard level.

[0020] In a specific embodiment, the process of executing step S1 may specifically include the following steps: A comprehensive test circuit connection is made based on standard measurement impedance, high-frequency coupling capacitor, digital bandpass filter, high-frequency signal amplifier and high-speed data acquisition card to form a high-frequency detection part; At least three disc-shaped UHF sensors, a UHF signal coupling system, a UHF bandpass filter, a UHF low noise amplifier and a UHF signal digitization module are selected and connected to form a UHF detection part; Selecting a synchronous trigger unit, connecting the high frequency detection part and the UHF detection part, to form a first detection circuit system; The first detection circuit system is subjected to impedance matching, and a disc-shaped UHF sensor is installed on the insulator flange of the GIS equipment and connected to the UHF detection part through a shielded cable to form a second detection circuit system; Performing electromagnetic shielding on the second detection circuit system, adding a grounded metal shielding layer outside the high-frequency detection part and the UHF detection part, connecting the metal shielding layer to the ground through a multi-point grounding method, and forming a dual detection circuit system; A sweep frequency signal is injected into the GIS equipment, and the response signal is collected through a dual detection circuit system to establish an equipment impedance characteristic model including the shell distributed inductance and the internal conductor distributed inductance.

[0021] Specifically, the electrical performance of each component inside the GIS equipment is effectively evaluated through standard measurement impedance, providing basic data reflecting electrical behavior. The function of the high-frequency coupling capacitor is to couple the test signal from the signal source to the inside of the GIS equipment without being interfered by external currents while maintaining the integrity of the signal. The digital bandpass filter filters out noise of non-target frequencies to ensure the clarity of the signal frequency band, and sets a bandpass frequency band to facilitate the detection of signals in a specific frequency band. The high-frequency signal amplifier is used to amplify the signal transmitted to the GIS equipment to ensure that the signal can maintain sufficient strength in various complex environments of the equipment, while the high-speed data acquisition card collects the signal in the equipment to ensure that the signal sampling and processing speed reaches the required accuracy and real-time performance. These components constitute the high-frequency detection part, which obtains the response signal of the GIS equipment and performs necessary signal processing on it. Through this high-frequency detection system, the high-frequency electromagnetic waves inside the equipment can be monitored, and then the status of the equipment can be evaluated. If the frequency characteristics of signal transmission are considered, a frequency response model is set using standard transmission line theory: in, is the impedance characteristic of the device, is the resistor, is the angular frequency of the signal, and are frequency-dependent inductance and capacitance, respectively. In this model, the electrical characteristics of GIS equipment in different working frequency bands are reflected by adjusting different measurement parameters. At least three disc-shaped UHF sensors are selected and connected with UHF signal coupling system, UHF bandpass filter, UHF low noise amplifier and UHF signal digitization module to form the UHF detection part. The function of UHF sensors is to detect discharge signals. They are arranged at different positions of the equipment so that partial discharge inside the equipment can be monitored in all directions. The function of UHF signal coupling system is to effectively guide UHF signals to signal amplification and digitization system. UHF bandpass filter is used to filter irrelevant frequencies to ensure that the signal is concentrated in the expected frequency range, while UHF low noise amplifier enhances signal strength while minimizing interference from external noise. UHF signal digitization module converts the amplified analog signal into digital signal for subsequent processing and analysis. UHF detection part can obtain signal characteristics related to partial discharge in real time. These signals contain electromagnetic characteristics of various discharge phenomena occurring in GIS equipment, such as instantaneous amplitude, spectrum, duration, etc. of discharge. Through signal analysis, these signals are pattern recognized to detect defects inside GIS equipment. In order to ensure that the high-frequency detection part and the UHF detection part can work together and improve the accuracy and timeliness of signal acquisition, the two parts are connected by selecting a synchronous trigger unit to form a first detection circuit system. The synchronous trigger unit is used to ensure the time synchronization of the two parts of the circuit, thereby avoiding data errors caused by asynchronous signal acquisition. Through the synchronous acquisition of the high-frequency detection part and the UHF detection part, the multi-dimensional characteristics of the signal are obtained at the same time, thereby improving the accuracy of detection. The first detection circuit system is impedance matched. Impedance mismatch will cause signal reflection and attenuation, affecting the transmission and acquisition of the signal. A matching network is added to the circuit to ensure the maximum transmission efficiency of the signal. Use the impedance matching formula to determine the appropriate circuit parameters. The impedance matching formula is: in, is the input impedance of the circuit, and They are the impedances of two circuit elements respectively. By adjusting these parameters, the optimal signal transmission can be achieved. The disc-shaped UHF sensor is installed at the insulator flange of the GIS equipment and connected to the UHF detection part through a shielded cable to form the second detection circuit system. By installing the sensor at the key part of the equipment, the internal discharge phenomenon of the equipment can be monitored in real time, and the shielded cable can effectively reduce external interference and improve the purity of the signal. In order to avoid the influence of electromagnetic interference on the test results, the second detection circuit system is electromagnetically shielded. The electromagnetic shielding layer is used to prevent the interference of external electromagnetic waves on the signal and reduce the external leakage of the signal. By adding a grounded metal shielding layer outside the high-frequency detection part and the UHF detection part, the external electromagnetic interference is effectively isolated and the signal stability is ensured. The metal shielding layer is connected to the earth by multi-point grounding, thereby improving the electromagnetic shielding effect and forming a dual detection circuit system. The sweep frequency signal is injected into the GIS equipment. The sweep frequency signal scans the electrical characteristics inside the equipment by gradually changing the frequency of the signal, thereby effectively identifying the defects in the equipment. The response signal collected by the dual detection circuit system is used to establish the equipment impedance characteristic model including the shell distributed inductance and the internal conductor distributed inductance. In this model, the shell distributed inductance and the internal conductor distributed inductance are key parameters for describing the electromagnetic characteristics of the internal and external parts of the GIS equipment, and can reflect the propagation of signals in the equipment. Based on these parameters, the corresponding electromagnetic wave propagation model is established to analyze the electrical behavior inside the equipment. For example, the inductance model of the equipment is given by the electromagnetic field theory: in, is the total inductance, and They are the inductance of the shell and the conductor respectively. According to the specific structure and measurement of the equipment, the inductance value is calculated and adjusted to further optimize the electrical characteristics of the equipment.

[0022] In a specific embodiment, the process of injecting a sweep frequency signal into the GIS device, collecting a response signal through a dual detection circuit system, and establishing a device impedance characteristic model including a shell distributed inductance and an internal conductor distributed inductance may specifically include the following steps: Inject the frequency sweep signal into the GIS device through the test port of the GIS device to perform frequency sweep; The response signals of the GIS equipment are synchronously collected through the dual detection circuit system, and the response signals collected by the high-frequency detection part are time-synchronized with the response signals collected by the UHF detection part to obtain the target response signal data set; Perform fast Fourier transform on the target response signal data set to obtain the frequency response characteristics of the GIS equipment; Based on the frequency response characteristics of GIS equipment, the distributed inductance of the GIS equipment shell, the distributed inductance of the internal conductor of the GIS equipment, the distributed capacitance between the internal conductor and the shell, the grounding resistance of the shell grounding point, the equivalent capacitance of the insulator and the equivalent parallel resistance of the insulator are extracted to obtain a distributed parameter set. Substitute the distributed parameter set into the Maxwell equations to calculate the propagation time delay, attenuation characteristics and reflection coefficient of the electromagnetic wave inside the GIS equipment, and obtain the electromagnetic propagation characteristic data; The electromagnetic propagation characteristic data are parameterized and an equipment impedance characteristic model describing the electromagnetic characteristics of GIS equipment is established.

[0023] Specifically, a signal injection system is designed. The system transmits a frequency sweep signal from an external signal source to the inside of the GIS device through a test port. The frequency sweep signal is a signal with a gradually changing frequency. Its frequency range covers the frequency range when the GIS device is working normally. The response of the device at different frequencies is captured by scanning the entire frequency range. After the frequency sweep signal is injected, the signal propagates along the various transmission paths of the device. Various electrical components inside the device (such as insulators, conductors, and metal shells, etc.) will affect the propagation of the signal. In order to ensure accurate signal acquisition and analysis, a dual detection circuit system is used to synchronously acquire the response signal of the GIS device. The dual detection circuit system includes a high-frequency detection part and a UHF detection part. The former detects the high-frequency signal in the device, and the latter captures the signal in the UHF range. By synchronizing the response signals of these two parts in time, it is ensured that the signal characteristics of different frequency bands are obtained at the same time, and the target response signal data set is obtained. The target response signal data set contains the electrical characteristics inside the device, such as the propagation path, attenuation, and possible reflection and interference of the signal. The target response signal data set is subjected to a fast Fourier transform to convert the signal from the time domain to the frequency domain, and the frequency response characteristics of the signal are obtained through frequency domain analysis. The frequency response characteristic reflects the device's ability to respond to signals of different frequencies, helping to identify the device's resonant frequency, filtering characteristics, and potential electromagnetic interference issues. By performing a fast Fourier transform on the target response signal, the frequency response obtained is expressed as: in, is the frequency response function, is the frequency content of the output signal, is the frequency content of the input signal, Indicates frequency. By analyzing the frequency response, we can understand the gain or attenuation of the device to a specific frequency signal and further determine the working status of the device. Based on the frequency response characteristics of the GIS device, electrical parameters are extracted. These parameters help describe the electromagnetic characteristics of the device. By analyzing the frequency response of the device, the distributed inductance of the GIS device housing and the distributed inductance of the internal conductor are calculated. Inductance is the impedance of a circuit element to current changes. It is closely related to the electrical characteristics of the device and can affect the response of the device at different frequencies. The distributed inductance model is expressed as: in, is the total inductance, and They are the distributed inductance of the shell and the conductor respectively. The distributed capacitance between the internal conductor and the shell in the computing device. Distributed capacitance refers to the capacitance formed by the electric field between the conductor and the conductor, and between the conductor and the ground, which will affect the transmission characteristics of the signal. The calculation of distributed capacitance is expressed by the following formula: in, is the frequency dependent capacitance, is the dielectric constant of the material, is the effective area of ​​the conductor, Is the distance between conductors. Calculate the grounding resistance of the shell grounding point, which is an important parameter of the equipment grounding system and directly affects the loop performance of the current. The calculation formula of grounding resistance is: in, is the ground resistance, is the resistivity of the grounding material, is the length of the grounding conductor, is the cross-sectional area of ​​the grounding conductor. Calculate the equivalent capacitance and equivalent parallel resistance of the insulator. The capacitance of the insulator affects the propagation of the signal, causing signal attenuation and distortion. Its equivalent capacitance and resistance are calculated by the following formula: in, is the equivalent capacitance of the insulator, is the equivalent resistance of the insulator, and are the dielectric constant and resistivity of the insulator material, and is the effective area of ​​the insulator, and are the thickness and length of the insulator respectively. Substitute the above distribution parameter set into the Maxwell equations to calculate the propagation characteristics of electromagnetic waves inside the GIS equipment. The Maxwell equations are the basic equations that describe the behavior of electromagnetic fields, including electric fields, magnetic fields, and their interactions. By solving these equations, information such as the propagation time delay, attenuation characteristics, and reflection coefficient of electromagnetic waves can be obtained. The general form of the Maxwell equations is as follows: in, is the electric displacement field, is the magnetic induction intensity, is the electric field strength, is the magnetic field strength, is the free charge density, is the current density. The electromagnetic propagation characteristic data are parameterized and converted into a form that is easy to analyze, and a device impedance characteristic model describing the electromagnetic characteristics of GIS equipment is established. Through parameterization, the frequency response curve of the equipment, the propagation delay of the electromagnetic wave, the reflection coefficient and the attenuation rate are obtained.

[0024] In a specific embodiment, the process of executing step S2 may specifically include the following steps: Apply rated operating voltage to GIS equipment, continuously monitor GIS equipment through dual detection circuit system, and obtain original discharge signal; The amplitude of the original discharge signal is calibrated based on the equipment impedance characteristic model, and the attenuation and distortion caused by signal propagation are compensated according to the electromagnetic propagation characteristic data of the GIS equipment to obtain the calibrated discharge signal. Preliminarily screening the calibrated discharge signal to select a valid discharge signal as the first multi-mode discharge signal; Performing empirical mode decomposition on the first multi-mode discharge signal to obtain a plurality of intrinsic mode function components; Collect environmental characteristic parameters of GIS equipment, including ambient temperature, ambient humidity, equipment temperature, equipment internal pressure, ambient electromagnetic field strength and equipment vibration parameters.

[0025] Specifically, the rated operating voltage is applied to the GIS equipment, which is the operating voltage of the equipment under standard working conditions. At this time, the electrical components inside the equipment work as expected, and partial discharge will occur under certain conditions. In order to capture these discharge signals in time, a dual detection circuit system is used for monitoring. The circuit system consists of two parts: a high-frequency detection part and a UHF detection part, which capture signals in different frequency bands respectively. In this way, the system monitors the signal changes inside the equipment and provides the original discharge signal data. The amplitude of the original discharge signal is calibrated based on the equipment impedance characteristic model. The impedance characteristic model of the equipment is used to compensate for the attenuation caused by signal propagation. The process of setting propagation attenuation is described by an exponential attenuation model: in, is the attenuation factor, is the signal amplitude after propagation, is the initial signal amplitude, is the frequency The attenuation coefficient under is the distance the signal propagates. According to this model, the signal is compensated and the calibrated signal is restored to truly reflect the discharge behavior in the device. The signal is screened to extract the effective discharge signal. The screening of effective signals is based on the amplitude, frequency and time domain characteristics of the signal. The effective discharge signal has a significant fluctuation characteristic, while the noise signal or irrelevant signal will be eliminated. Use an amplitude threshold To filter: in, is the time domain signal, Is a valid signal. The signal amplitude is greater than the threshold When , the signal is considered to be an effective discharge signal, otherwise it is a noise signal. Through this step, the first multi-mode discharge signal is obtained. The first multi-mode discharge signal is subjected to empirical mode decomposition, and the nonlinear and non-stationary signal is decomposed into several intrinsic mode function (IMF) components. Each IMF component represents the characteristics of the signal at a certain frequency or time scale. After the empirical mode decomposition process, the original signal is decomposed into multiple IMF components and a residual term : in, It is The eigenmode functions, is the residual term, is the number of IMFs obtained by decomposition. The results of empirical mode decomposition help analyze the oscillation characteristics of the signal in different frequency bands. At the same time, the environmental characteristic parameters of the device are collected synchronously. Environmental characteristic parameters include ambient temperature, humidity, device temperature, device internal pressure, ambient electromagnetic field strength, and device vibration, which will affect the discharge behavior of the device.

[0026] In a specific embodiment, the process of performing the step of performing empirical mode decomposition on the first multi-mode discharge signal to obtain a plurality of intrinsic mode function components may specifically include the following steps: Preprocessing the first multi-mode discharge signal to obtain a preprocessed discharge signal; The empirical mode decomposition algorithm is applied to the preprocessed discharge signal. The local extreme points of the preprocessed discharge signal are extracted through an iterative screening process, and the upper and lower envelopes are constructed. The mean of the envelope is calculated as the mean curve to obtain the preliminary decomposition result. Perform termination judgment on the preliminary decomposition result to obtain M initial eigenmode function components and a residual term; Performing Hilbert transform on each initial intrinsic mode function component, calculating instantaneous frequency, spectrum entropy, energy ratio and signal complexity, and forming a characteristic parameter set of the initial intrinsic mode function component; Based on the device impedance characteristic model, the propagation mode matching degree is calculated for the initial eigenmode function components, and the propagation mode matching degree result corresponding to each initial eigenmode function component is identified; According to the propagation mode matching result, the initial intrinsic mode function components are grouped and reconstructed to obtain a plurality of separated intrinsic mode function components.

[0027] Specifically, the first multi-mode discharge signal is denoised and filtered by wavelet transform denoising, bandpass filtering and empirical mode decomposition preprocessing. In the time domain, the filtered signal is expressed as: in, represents the preprocessed signal, is the original signal, and are Fourier transform and inverse Fourier transform, respectively. is the transfer function of the bandpass filter. By filtering, the effective signal components are retained and high-frequency noise is removed, making the signal more stable. The empirical mode decomposition algorithm is applied to process the preprocessed signal, extract the local extreme points of the signal, construct the upper and lower envelopes, and calculate the mean of the envelopes as the mean curve. The upper envelope and the lower envelope are interpolated from the local maximum and local minimum of the signal, respectively. The mean curve is calculated as follows: in, is the upper envelope, is the lower envelope, is the mean curve. Then remove the mean curve from the signal and get the new signal: This process is repeated until the signal meets the definition of IMF, that is, its envelope mean is close to zero. The signal is decomposed into multiple intrinsic mode function (IMF) components and a residual term: in, is the number of IMFs obtained by decomposition, Indicates IMF components, is the residual signal. For the number of IMF components , judged by the following termination conditions: in, The energy threshold is set to ensure that the IMF component covers the main components of the signal without introducing excess noise. Hilbert transform is performed on each IMF component to calculate the instantaneous frequency, spectral entropy, energy ratio and signal complexity. The mathematical definition of Hilbert transform is: in, To analyze the signal, For signal The result after Hilbert transform is: is an imaginary unit. By analyzing the signal, calculate the instantaneous frequency: in, To resolve the phase of the signal, is the instantaneous amplitude. For spectral entropy, it is defined as follows: in, is the spectrum entropy, The signal in the frequency domain The probability density of frequency components. Spectral entropy measures the complexity of the signal's spectral distribution. The larger the value, the more dispersed the frequency distribution, and the smaller the value, the signal is concentrated on a few frequencies. The energy ratio is calculated as follows: in, Indicates The energy ratio of the IMF components measures the energy proportion of the component in the entire signal. This indicates that the IMF contributes greatly to the signal. The calculation of signal complexity is based on approximate entropy or sample entropy, expressed as: in, Measures the irregularity of the signal. The larger the value, the more complex the signal. Based on the impedance characteristic model of the device, the propagation mode matching of each IMF component is calculated. The calculation of propagation mode matching depends on the impedance characteristics and electromagnetic propagation characteristics of the GIS device, and is defined as follows: in, Indicates and The matching degree of the IMF components is calculated. The higher the matching degree, the similar their propagation modes are. According to the matching results, the IMF components are grouped and reconstructed. For the IMF components with higher matching degrees, they are grouped into the same group for reconstruction to remove irrelevant components and extract the main propagation mode. The reconstructed signal is expressed as: in, is the reconstructed component set, is the new signal after grouping. Through this process, the useful signal pattern is extracted and the redundant interference items are removed.

[0028] In a specific embodiment, the process of executing step S3 may specifically include the following steps: Separate the features of multiple intrinsic mode function components and environmental characteristic parameters to form a time domain feature set and a frequency domain feature set; A variable-length sequence feature extraction module is constructed based on the time-domain feature set. The variable-length sequence feature extraction module includes a multi-scale convolution layer, an attention pooling layer, and a greedy sequence selector. A Transformer encoder network is constructed based on the output of the frequency domain feature set and the variable-length sequence feature extraction module. The Transformer encoder network includes a position encoding layer, a multi-head self-attention layer, and a feedforward neural network. A Bayesian network inference module is constructed based on the output features of the Transformer encoder network. The Bayesian network inference module includes a conditional probability table, a probability propagation algorithm, and a posterior probability calculation unit. The output of the Bayesian network inference module is used as a reward signal to build a proximal policy optimization training system, which includes a policy network, a value network, and an optimizer. The output of the proximal strategy optimization training system is verified and the model is selected through K-fold cross validation to obtain the discharge signal recognition model.

[0029] Specifically, multiple intrinsic mode function components and environmental characteristic parameters are separated to form time domain feature sets and frequency domain feature sets. Time domain features include instantaneous amplitude, mean, standard deviation, skewness, kurtosis, etc., while frequency domain features are composed of power spectrum density, center frequency, bandwidth, spectral entropy and other information obtained by Fourier transform or wavelet transform. Environmental characteristic parameters such as temperature, humidity, pressure, electromagnetic interference, etc. are included in the feature set to supplement the interpretation ability of signal characteristics and improve the adaptability of the model. Assume that the original signal is , time domain characteristics Calculated by the following statistics: in, is the mean, is the standard deviation, is the skewness, is the total number of samples. At the same time, the frequency domain characteristics of the signal are transformed into power spectrum density through Fourier transform : in, Represents Fourier transform. Based on the time domain feature set, a variable length sequence feature extraction module is constructed, which adapts to the change of discharge signal length under different behavior modes. The variable length sequence feature extraction module consists of a multi-scale convolution layer, an attention pooling layer, and a greedy sequence selector. Among them, the multi-scale convolution layer is used to extract signal patterns on different time scales, and its output is expressed as: in, It is The signal after the convolution layer, is the convolution kernel parameter, is the convolution window size. The attention pooling layer calculates the attention weights at different time steps To perform weighting: in, is the trainable attention weight. The greedy sequence selector is used to select the most representative time series fragments to reduce computational complexity and remove redundant information. Based on the output of the frequency domain feature set and the variable length sequence feature extraction module, a Transformer encoder network is constructed. The network includes a position encoding layer, a multi-head self-attention layer, and a feedforward neural network. Among them, the position encoding layer is used to retain the time series information so that the model can understand the temporal dependency of the signal. The position encoding is defined using sine and cosine functions: in, and It is The position encoding of the time step, is the hidden layer dimension of the model. The multi-head self-attention layer calculates the attention weight matrix Perform global feature modeling: in, are the query, key, and value matrices of the input sequence, respectively, is the feature dimension. The feedforward neural network is used for nonlinear transformation and is defined as follows: in, is a trainable parameter, is the bias term. Based on the output features of the Transformer encoder network, a Bayesian network inference module is constructed. This module includes a conditional probability table, a probability propagation algorithm, and a posterior probability calculation unit. Among them, the conditional probability table is used to store the joint probability distribution between the discharge mode and the environmental parameters. Assuming that the discharge type and the eigenvector Satisfies Bayes' theorem: in, is the posterior probability, is the likelihood probability, is the prior probability, is the evidence item. The probability propagation algorithm derives the most likely discharge category by calculating the marginal probability and the maximum a posteriori probability. The posterior probability calculation unit is used to dynamically update the confidence of the model so that the model can be adaptively adjusted when facing new data. In order to optimize the discharge signal recognition model and improve the generalization ability of the model, the output result of the Bayesian network inference module is used as a reward signal to construct a proximal policy optimization (PPO) training system. The PPO training system continuously optimizes the decision-making ability of the model through reinforcement learning, and is mainly composed of a policy network, a value network, and an optimizer. The policy network is used to generate classification decisions for discharge signals, and its parameter update adopts the following objective function: in, are policy network parameters, For advantage estimation, is the importance sampling ratio, is the cutoff coefficient. The value network is used to evaluate the pros and cons of the current strategy, and its optimization goal is to minimize the mean square error: in, is the value estimate of the current state, is the target value. The optimizer uses the gradient descent method to adjust the network parameters so that the classification ability of the model is continuously improved. The output of the proximal strategy optimization training system is verified and the model is selected through K-fold cross validation. The process of K-fold cross validation is to divide the data set into Each time, select one of them as the test set and the rest as the training set. Repeat The training and testing are repeated, and the average value is taken as the evaluation index of the model: in, It is Through this process, the optimal discharge signal recognition model is selected and its generalization ability is ensured.

[0030] In a specific embodiment, the process of executing step S4 may specifically include the following steps: Performing online monitoring on the GIS equipment, collecting the second multi-mode discharge signal through the dual detection circuit system, and preprocessing and empirical mode decomposition of the second multi-mode discharge signal to obtain the eigenmode function component for testing; The test intrinsic mode function component is input into the discharge signal recognition model, the time-frequency features are extracted by the variable-length sequence feature extraction module, the features are enhanced by the multi-head self-attention layer, and the probability distribution of each discharge type is calculated by the Bayesian network inference module to obtain the discharge type recognition result; Based on the discharge signals collected by at least three disc-shaped UHF sensors of the UHF detection part, the time when the discharge signals arrive at each sensor is recorded, and the time difference matrix between every two sensors is calculated; Substitute the time difference matrix into the hyperbolic positioning equations, use the Taylor series expansion method to linearize the nonlinear equations, and iteratively solve the equations using the weighted least squares method to obtain the three-dimensional coordinates of the internal defects. Determine the defect type based on the discharge type identification result. The defect types include metal suspended particles, insulator surface discharge, conductor tip discharge, insulator internal air gap discharge and poor contact discharge. Based on the three-dimensional coordinates of the defect and the defect type, the discharge repetition rate, discharge phase stability, discharge amplitude trend and discharge energy density are calculated, and the hazard level is obtained through weighted fusion.

[0031] Specifically, the GIS equipment is monitored online, and the second multi-mode discharge signal is collected by using a dual detection circuit system, and preprocessed and empirical mode decomposed to obtain the eigenmode function components for testing. During the monitoring process, the high-frequency detection part and the UHF detection part synchronously collect the discharge signal, and remove the noise through a bandpass filter to keep a high signal-to-noise ratio in the frequency range of interest. Assume that the original discharge signal is , and its preprocessing process is expressed as: in, is the filtered signal, and are Fourier transform and inverse Fourier transform, respectively. is the transfer function of the bandpass filter. Perform empirical mode decomposition on the preprocessed signal to extract multiple intrinsic mode function components. Empirical mode decomposition iteratively extracts the local extreme points of the signal and constructs the upper and lower envelopes to calculate the mean curve: in, and are the upper envelope and the lower envelope respectively, is the mean curve. By removing the mean curve and iterative processing, we get Eigenmode function components and the residual : in, It is The eigenmode functions, Represents the number of IMF decomposition layers. The test intrinsic mode function components are input into the discharge signal recognition model. The model extracts time-frequency features through a variable-length sequence feature extraction module, in which the multi-scale convolution layer is used to capture signal patterns on different time scales. The calculation formula is: in, It is The signal after the convolution layer, is the convolution kernel parameter, is the convolution window size. The attention pooling layer calculates the attention weights at different time steps: in, is a trainable attention parameter. Multi-head self-attention layers are used to enhance features so that features at different time steps can be correlated with each other. The calculation formula is as follows: in, represent the query matrix, key matrix and value matrix of the input features respectively, is the feature dimension. The Bayesian network inference module is used to calculate the probability distribution of the discharge signal belonging to various defects. The Bayesian theorem is expressed as follows: in, Represents a given observation When the discharge type The posterior probability of is the likelihood probability, is the prior probability, is a normalized term. The most likely discharge category is calculated through the probability propagation algorithm. In order to determine the location of the defect, the discharge signal collected by at least three disc-type UHF sensors in the UHF detection part is used to record the time when the signal arrives at each sensor, and the time difference matrix between the sensors is calculated: in, For sensor and The time difference between and Substitute the time difference matrix into the hyperbolic positioning equations, use the Taylor series expansion method to linearize the nonlinear equations, and use the weighted least squares method to iteratively solve them, and finally get the three-dimensional coordinates of the defect. : in, are the known coordinates of the sensor, is the signal propagation speed, is the time it takes for the signal to propagate to an unknown point. The defect type is determined based on the discharge type identification results, which mainly include metal suspended particles, insulator surface discharge, conductor tip discharge, insulator internal air gap discharge, and poor contact discharge. Different types of discharges correspond to different electromagnetic wave characteristics. For example, metal suspended particle discharges usually manifest as high-frequency short-time shock waves, while insulator internal air gap discharges have relatively stable harmonic characteristics. Based on the three-dimensional coordinates of the defect and the defect type, the discharge repetition rate, discharge phase stability, discharge amplitude trend, and discharge energy density are calculated. The discharge repetition rate indicates the frequency of discharge, which can be calculated by the following formula: in, is the discharge repetition rate, is the number of discharges during the observation time, To observe the time discharge phase stability, the phase distribution of the discharge is measured and defined as follows: in, For the The phase angle of the discharge. The discharge amplitude trend is measured by calculating the rate of change of the discharge amplitude in the time series, expressed as: in, For the The discharge energy density indicates the discharge energy per unit time: By weighted integration of the above parameters, the comprehensive hazard level is calculated: in, are the weights of different parameters.

[0032] The above describes the partial discharge detection method of the fully enclosed vacuum environmentally friendly GIS equipment in the embodiment of the present invention. The following describes the partial discharge detection device of the fully enclosed vacuum environmentally friendly GIS equipment in the embodiment of the present invention. Figure 2 , an embodiment of the partial discharge detection device of the fully enclosed vacuum environmentally friendly GIS equipment in the embodiment of the present invention includes: The detection module is used to construct a dual detection circuit system for GIS equipment and establish an equipment impedance characteristic model including the distributed inductance of the shell and the distributed inductance of the internal conductor; A decomposition module is used to collect the first multi-mode discharge signal of the GIS equipment in a vacuum insulation environment and perform empirical mode decomposition based on the equipment impedance characteristic model to obtain multiple intrinsic mode function components and simultaneously record environmental characteristic parameters; A construction module, used for constructing a discharge signal recognition model based on a plurality of intrinsic mode function components and environmental characteristic parameters; The identification module is used to identify the second multi-mode discharge signal of the GIS equipment through the discharge signal identification model, and determine the three-dimensional coordinates and hazard level of the internal defects of the GIS equipment.

[0033] Through the cooperation of the above-mentioned components, the dual detection circuit system combining high-frequency detection and UHF detection constructed by the present invention realizes full-spectrum monitoring of discharge signals, which can not only quantitatively measure the amount of discharge charge, but also accurately locate the discharge source, effectively solving the problem of incomplete information of a single detection method. Secondly, by injecting a sweep frequency signal into the GIS device and collecting the response signal, an equipment impedance characteristic model including the shell distributed inductance and the internal conductor distributed inductance is established, accurately characterizing the electromagnetic wave propagation characteristics inside the device. The discharge signal is processed by the empirical mode decomposition technology to effectively separate the multiple discharge types that exist at the same time, overcoming the limitations of traditional single mode recognition. The method synchronously records the environmental characteristic parameters and associates them with the discharge signal, establishes a "discharge signal-environmental parameter" mapping relationship, and enables the detection system to adaptively adjust the recognition strategy according to environmental changes. The constructed PTB model integrates a variable-length sequence feature extraction module, a Transformer encoder network, and a Bayesian network reasoning module, and is trained through a proximal strategy optimization algorithm to realize intelligent recognition of discharge characteristics, improve computational efficiency, and enhance anti-interference ability. Based on the UHF sensor array and hyperbolic positioning algorithm, combined with the PTB model recognition results, the three-dimensional coordinates of internal defects of the equipment can be accurately calculated, realizing a comprehensive assessment of defect type and hazard level.

[0034] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0035] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0036] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A partial discharge detection method for a fully enclosed vacuum environmentally friendly GIS device, characterized in that: The method comprises: Construct a dual detection circuit system for GIS equipment and establish an equipment impedance characteristic model including the distributed inductance of the shell and the distributed inductance of the internal conductor; Based on the device impedance characteristic model, the first multi-mode discharge signal of the GIS device in a vacuum insulation environment is collected and empirical mode decomposition is performed to obtain multiple intrinsic mode function components, and environmental characteristic parameters are simultaneously recorded; Constructing a discharge signal recognition model based on the multiple eigenmode function components and the environmental characteristic parameters; The second multi-mode discharge signal of the GIS device is identified by the discharge signal identification model to determine the three-dimensional coordinates and hazard level of the internal defects of the GIS device.

2. The partial discharge detection method of the fully enclosed vacuum environmentally friendly GIS equipment according to claim 1 is characterized in that: The dual detection circuit system of the GIS equipment is constructed, and the equipment impedance characteristic model including the shell distributed inductance and the internal conductor distributed inductance is established, including: A comprehensive test circuit connection is made based on standard measurement impedance, high-frequency coupling capacitor, digital bandpass filter, high-frequency signal amplifier and high-speed data acquisition card to form a high-frequency detection part; At least three disc-shaped UHF sensors, a UHF signal coupling system, a UHF bandpass filter, a UHF low noise amplifier and a UHF signal digitization module are selected and connected to form a UHF detection part; Selecting a synchronous trigger unit, connecting the high frequency detection part and the UHF detection part to form a first detection circuit system; Performing impedance matching processing on the first detection circuit system, and installing the disc-shaped UHF sensor on the insulator flange of the GIS equipment, and connecting it to the UHF detection part through a shielded cable to form a second detection circuit system; Performing electromagnetic shielding on the second detection circuit system, adding a grounded metal shielding layer outside the high-frequency detection part and the UHF detection part, and connecting the metal shielding layer to the ground through a multi-point grounding method to form a dual detection circuit system; A sweep frequency signal is injected into the GIS device, a response signal is collected through the dual detection circuit system, and a device impedance characteristic model including the shell distributed inductance and the internal conductor distributed inductance is established.

3. The partial discharge detection method of the fully enclosed vacuum environmentally friendly GIS equipment according to claim 2 is characterized in that: The step of injecting a frequency sweep signal into the GIS device, collecting a response signal through the dual detection circuit system, and establishing a device impedance characteristic model including a shell distributed inductance and an internal conductor distributed inductance includes: Injecting the frequency sweep signal into the GIS device through the test port of the GIS device to perform frequency scanning; The response signal of the GIS device is synchronously collected by the dual detection circuit system, and the response signal collected by the high-frequency detection part is time-synchronized with the response signal collected by the UHF detection part to obtain a target response signal data set; Performing a fast Fourier transform on the target response signal data set to obtain a frequency response characteristic of the GIS device; Based on the frequency response characteristics of the GIS equipment, the distributed inductance of the GIS equipment shell, the distributed inductance of the internal conductor of the GIS equipment, the distributed capacitance between the internal conductor and the shell, the grounding resistance of the shell grounding point, the equivalent capacitance of the insulator and the equivalent parallel resistance of the insulator are extracted to obtain a distributed parameter set; Substituting the distributed parameter set into the Maxwell equations, calculating the propagation time delay, attenuation characteristics and reflection coefficient of the electromagnetic wave inside the GIS device, and obtaining electromagnetic propagation characteristic data; The electromagnetic propagation characteristic data is parameterized to establish an equipment impedance characteristic model that describes the electromagnetic characteristics of the GIS equipment.

4. The method for detecting partial discharge of a fully enclosed vacuum environment-friendly GIS device according to claim 1 is characterized in that: Based on the device impedance characteristic model, the first multi-mode discharge signal of the GIS device in a vacuum insulation environment is collected and empirical mode decomposition is performed to obtain multiple intrinsic mode function components, and environmental characteristic parameters are simultaneously recorded, including: Applying a rated operating voltage to the GIS device, continuously monitoring the GIS device through the dual detection circuit system, and obtaining an original discharge signal; The amplitude of the original discharge signal is calibrated based on the device impedance characteristic model, and the attenuation and distortion caused by signal propagation are compensated according to the electromagnetic propagation characteristic data of the GIS device to obtain a calibrated discharge signal; Preliminarily screening the calibrated discharge signal to select a valid discharge signal as the first multi-mode discharge signal; Performing empirical mode decomposition on the first multi-mode discharge signal to obtain a plurality of intrinsic mode function components; The environmental characteristic parameters of the GIS equipment are collected, and the environmental characteristic parameters include environmental temperature, environmental humidity, equipment temperature, equipment internal pressure, environmental electromagnetic field strength and equipment vibration parameters.

5. The partial discharge detection method of the fully enclosed vacuum environmentally friendly GIS equipment according to claim 4 is characterized in that: The step of performing empirical mode decomposition on the first multi-mode discharge signal to obtain a plurality of intrinsic mode function components includes: Preprocessing the first multi-mode discharge signal to obtain a preprocessed discharge signal; Applying an empirical mode decomposition algorithm to the preprocessed discharge signal, extracting local extreme points of the preprocessed discharge signal through an iterative screening process and constructing upper and lower envelopes, calculating the mean of the envelopes as a mean curve, and obtaining a preliminary decomposition result; Performing termination judgment on the preliminary decomposition result to obtain M initial eigenmode function components and a residual term; Performing Hilbert transform on each of the initial intrinsic mode function components, calculating instantaneous frequency, spectrum entropy, energy ratio and signal complexity, and forming a characteristic parameter set of the initial intrinsic mode function components; Based on the device impedance characteristic model, calculating the propagation mode matching degree for the initial eigenmode function components, and identifying the propagation mode matching degree result corresponding to each initial eigenmode function component; According to the propagation mode matching result, the initial intrinsic mode function components are grouped and reconstructed to obtain a plurality of separated intrinsic mode function components.

6. The partial discharge detection method of the fully enclosed vacuum environmentally friendly GIS equipment according to claim 1 is characterized in that: The step of constructing a discharge signal recognition model based on the multiple eigenmode function components and the environmental characteristic parameters includes: Separating the plurality of intrinsic mode function components and the environmental characteristic parameters to form a time domain feature set and a frequency domain feature set; Constructing a variable-length sequence feature extraction module based on the time-domain feature set, wherein the variable-length sequence feature extraction module includes a multi-scale convolution layer, an attention pooling layer, and a greedy sequence selector; Constructing a Transformer encoder network based on the frequency domain feature set and the output of the variable-length sequence feature extraction module, wherein the Transformer encoder network includes a position encoding layer, a multi-head self-attention layer, and a feedforward neural network; Constructing a Bayesian network reasoning module based on the output features of the Transformer encoder network, wherein the Bayesian network reasoning module includes a conditional probability table, a probability propagation algorithm, and a posterior probability calculation unit; Using the output result of the Bayesian network reasoning module as a reward signal, constructing a proximal strategy optimization training system, wherein the proximal strategy optimization training system includes a strategy network, a value network, and an optimizer; The output of the proximal strategy optimization training system is verified and a model is selected through K-fold cross validation to obtain a discharge signal recognition model.

7. The partial discharge detection method of the fully enclosed vacuum environmentally friendly GIS equipment according to claim 2 is characterized in that: The step of identifying the second multi-mode discharge signal of the GIS device by the discharge signal identification model to determine the three-dimensional coordinates and hazard level of the internal defects of the GIS device includes: Performing online monitoring on the GIS equipment, collecting a second multi-mode discharge signal through the dual detection circuit system, and performing preprocessing and empirical mode decomposition on the second multi-mode discharge signal to obtain an intrinsic mode function component for testing; The test eigenmode function component is input into the discharge signal recognition model, the time-frequency features are extracted by the variable-length sequence feature extraction module, the features are enhanced by the multi-head self-attention layer, and the probability distribution of each discharge type is calculated by the Bayesian network reasoning module to obtain the discharge type recognition result; Based on the discharge signals collected by at least three disc-shaped UHF sensors of the UHF detection part, the time when the discharge signals arrive at each sensor is recorded, and the time difference matrix between every two sensors is calculated; Substituting the time difference matrix into the hyperbolic positioning equation group, linearizing the nonlinear equation group by Taylor series expansion method, iteratively solving the equation group by weighted least square method, and obtaining the three-dimensional coordinates of the internal defect; Determining the defect type according to the discharge type identification result, wherein the defect type includes metal suspended particles, insulator surface discharge, conductor tip discharge, insulator internal air gap discharge and poor contact discharge; Based on the three-dimensional coordinates of the defect and the defect type, the discharge repetition rate, the discharge phase stability, the discharge amplitude trend and the discharge energy density are calculated, and the hazard level is obtained through weighted fusion.

8. A partial discharge detection device for a fully enclosed vacuum environmentally friendly GIS device, characterized in that: Used to perform the partial discharge detection method of the fully enclosed vacuum environmentally friendly GIS equipment as described in any one of claims 1 to 7, the partial discharge detection device of the fully enclosed vacuum environmentally friendly GIS equipment comprises: The detection module is used to construct a dual detection circuit system for GIS equipment and establish an equipment impedance characteristic model including the distributed inductance of the shell and the distributed inductance of the internal conductor; A decomposition module is used to collect the first multi-mode discharge signal of the GIS device in a vacuum insulation environment based on the device impedance characteristic model and perform empirical mode decomposition to obtain multiple intrinsic mode function components and simultaneously record environmental characteristic parameters; A construction module, used for constructing a discharge signal recognition model based on the multiple eigenmode function components and the environmental characteristic parameters; The identification module is used to identify the second multi-mode discharge signal of the GIS device through the discharge signal identification model, and determine the three-dimensional coordinates and hazard level of the internal defects of the GIS device.

Images