GIS combined electrical apparatus operation on-line detection method and system

Abstract

This invention discloses an online detection method for GIS combined electrical appliances, relating to the technical field, and includes the following steps: acquiring the original time-domain signal of partial discharge, performing signal enhancement processing on the original time-domain signal to extract weak discharge pulse signals, and generating enhanced discharge pulse signals; constructing a holographic time-frequency matrix map based on the enhanced discharge pulse signals; using the holographic time-frequency matrix map as input, identifying the type of weak discharge based on a weak discharge neural network recognition model; and acquiring the spatial coordinates of the discharge source based on the recognition results and sensor array data, and outputting the discharge type and positioning results to the monitoring terminal. This invention significantly improves the identification capability, noise resistance, and positioning accuracy of the GIS combined electrical appliance online detection system for weak discharges by integrating a holographic time-frequency multi-scale analysis algorithm with a weak discharge neural network recognition mechanism.

Description

Technical Field

[0001] This invention relates to the field of GIS online monitoring technology, and more specifically, to a method and system for online monitoring of the operation of GIS combined electrical appliances. Background Technology

[0002] Gas-insulated metal-enclosed switchgear (GIS) is a critical switching device in high-voltage power transmission and transformation systems, and its operational reliability is of great significance to the safety and stability of the entire power grid. Due to its compact structure and strong enclosure, GIS is highly susceptible to serious accidents such as breakdown and explosion if insulation defects such as partial discharge (PD) occur inside. Therefore, developing online monitoring of the operational status of GIS equipment and early warning of insulation degradation has become an important technical direction for condition-based maintenance and intelligent operation and maintenance of power equipment.

[0003] Currently, online partial discharge monitoring technology has become an important means of GIS condition assessment, especially detection methods based on ultra-high frequency (UHF) signals. Due to their non-contact nature, strong anti-interference capabilities, and ability to detect early insulation degradation, they have been widely deployed in engineering applications. Although the UHF detection method technology system is relatively mature, the following key technical bottlenecks still exist in practical engineering applications:

[0004] Current methods have limited ability to detect extremely weak partial discharges. Traditional filtering and thresholding methods struggle to distinguish between effective signals and background noise, leading to frequent false positives and false negatives. Furthermore, classification based on single parameters or statistical features is susceptible to changes in environmental conditions and equipment type, lacking versatility and adaptability, and making it difficult to accurately identify typical defect morphologies of weak discharge types.

[0005] To address the above problems, this invention proposes a solution. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an online detection method and system for GIS combined electrical appliances. By integrating a holographic time-frequency multi-scale analysis algorithm with a weak discharge neural network identification mechanism, the online detection system for GIS combined electrical appliances significantly improves its ability to identify weak discharges, its noise resistance robustness, and its positioning accuracy.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] In the first aspect, this application provides an online detection method for GIS combined electrical appliances. The method includes: acquiring the original time-domain signal of partial discharge, performing signal enhancement processing on the original time-domain signal, extracting weak discharge pulse signals, and generating enhanced discharge pulse signals; constructing a holographic time-frequency matrix map based on the enhanced discharge pulse signals; using the holographic time-frequency matrix map as input, identifying the type of weak discharge based on a weak discharge neural network recognition model; and acquiring the spatial coordinates of the discharge source based on the recognition results and sensor array data, and outputting the discharge type and positioning results to the monitoring terminal.

[0009] In one embodiment, signal enhancement processing is performed on the original time-domain signal to extract weak discharge pulse signals and generate enhanced discharge pulse signals. Specifically, the original time-domain signal is decomposed into multi-level wavelet packet decomposition to obtain multi-channel sub-frequency band components; each sub-frequency band component is sorted using the entropy weight method to identify the target frequency band signal component that highlights the pulse energy concentration; an atom dictionary is constructed based on the temporal structure characteristics of the target frequency band components; the atom dictionary is used as the basis for sparse modeling, and the orthogonal matching pursuit algorithm is used to sparsely decompose the target frequency band signal components to obtain sparse atoms and their corresponding sparse coefficients; from the sparse atoms, they are sorted according to the matching value between each atom and the current signal residual; atoms with matching values ​​greater than a preset threshold are selected as a subset of atoms with high matching degree with the weak discharge pulse characteristics; signal reconstruction is performed based on the subset of atoms with high matching degree and their sparse coefficients to generate a sparse enhanced signal; phase consistency recovery and temporal reconstruction are performed on the sparse enhanced signal to output the enhanced discharge pulse signal.

[0010] In one embodiment, each sub-band component is sorted using the entropy weight method to identify the target frequency band signal component that highlights the pulse energy concentration. Specifically, the energy value of each sub-band component is calculated, and a sub-band energy vector is constructed. The sub-band energy vector is normalized to obtain the sub-band energy probability distribution, and the energy entropy value of each sub-band is calculated based on information entropy. The energy entropy value of each sub-band is converted into a weight, and the pulse energy concentration factor is calculated in combination with the energy value to construct a concentration vector. The concentration vector is sorted in descending order, and the sub-band components corresponding to the first few concentration vectors are selected as the target frequency band signal components.

[0011] In one embodiment, an orthogonal matching pursuit algorithm is used to sparsely decompose the target frequency band signal components to obtain sparse atoms and their corresponding sparse coefficients. Specifically, the target frequency band signal components are used as initial residuals. In each iteration, the atom corresponding to the largest absolute value of the inner product between the current residual and all atoms in the dictionary is selected, and the selected atom is added to the set of selected atoms. After each iteration, based on the current set of selected atoms, the sparse coefficients corresponding to the selected atoms are obtained by the least squares method. After the preset termination condition is met, a sparse atom subset consisting of a small number of atoms and its corresponding sparse coefficients are output.

[0012] In one embodiment, a holographic time-frequency matrix map is constructed based on the enhanced discharge pulse signal. Specifically, the enhanced discharge pulse signal is windowed and subjected to a short-time Fourier transform to obtain an initial time-frequency distribution matrix; a discrete wavelet packet transform is performed on the enhanced discharge pulse signal to decompose the signal layer by layer to a preset number of wavelet packet decomposition layers, obtaining several frequency band sub-signals; energy mapping is performed on each frequency band sub-signal to construct a wavelet packet frequency band energy matrix; the initial time-frequency distribution matrix and the wavelet packet frequency band energy matrix are spatially aligned and feature fusion is performed to generate a holographic time-frequency matrix map.

[0013] In one embodiment, a holographic time-frequency matrix spectrum is used as input to identify weak discharge types based on a weak discharge neural network recognition model. Specifically, the holographic time-frequency matrix spectrum is subjected to dimensionality reduction using a local preservation projection method to extract low-dimensional feature representations with high information entropy. The low-dimensional feature representations are then temporally reconstructed and expanded to construct a tensor containing temporal correlation and spectral structure. Based on this tensor, a weak discharge neural network recognition model is constructed, comprising a shallow convolutional feature extraction network, a deep residual feature extraction module, an attention mechanism module, and a fully connected neural network classification layer. A labeled weak discharge sample spectrum dataset is acquired to train the model using supervised learning. The trained weak discharge neural network recognition model analyzes the holographic time-frequency matrix spectrum to be identified and outputs corresponding weak discharge type classification results, including a probability value for each type. The output probability value is used to determine whether the weak discharge type is identifiable; if it exceeds a threshold, the final identification type is output.

[0014] In one embodiment, a trained weak discharge neural network recognition model analyzes the holographic time-frequency matrix spectrum to be identified and outputs the corresponding weak discharge type classification result. Specifically, the tensor is input into a shallow convolutional feature extraction network, and features are extracted through several convolutional and pooling layers to form a first-stage partial discharge feature map tensor. The partial discharge feature map tensor is input into a deep residual feature extraction module, and a deep tensor containing time-frequency semantic features is output. The deep tensor is input into an attention mechanism module, and a weighted feature map tensor is output. The weighted feature map tensor is subjected to global average pooling and feature flattening to transform it into a one-dimensional feature vector. The one-dimensional feature vector is input into a fully connected neural network classification layer, and a probability value vector for each weak discharge type is output. The probability value vector is normalized based on the Softmax function so that the sum of the probabilities of each category is 1. Finally, the weak discharge type corresponding to the highest probability is output as the recognition result.

[0015] In one embodiment, based on the identification results and sensor array data, the spatial coordinates of the discharge source are obtained, and the discharge type and location results are output to the monitoring terminal. Specifically, the discharge occurrence time is obtained based on the model's identification results, and sensor array data for the corresponding time period is retrieved. The sensor array data includes spatial layout coordinate information, discharge response waveforms of each channel, and their absolute timestamps. Using the timestamps, the arrival time difference between each channel is calculated based on a generalized cross-correlation algorithm to form a time difference matrix. Based on the time difference matrix and combined with the spatial layout coordinate information of the sensor array, the spatial coordinate position of the discharge source is obtained using a nonlinear least squares method. The discharge type identification results and the spatial coordinate position of the discharge source are fused to form a structured result set of the discharge event. The structured result set is output to the power equipment status monitoring terminal through a communication module.

[0016] Secondly, this application provides an online monitoring system for the operation of GIS combined electrical appliances. This system includes a signal enhancement module, a holographic time-frequency matrix map construction module, a weak discharge type identification module, and a discharge event output module, with connections between the modules.

[0017] The signal enhancement module is used to acquire the original time-domain signal of partial discharge, perform signal enhancement processing on the original time-domain signal, extract the weak discharge pulse signal, and generate the enhanced discharge pulse signal.

[0018] The holographic time-frequency matrix map construction module is used to construct a holographic time-frequency matrix map based on the enhanced discharge pulse signal;

[0019] The weak discharge type identification module is used to identify the weak discharge type by taking the holographic time-frequency matrix spectrum as input and based on the weak discharge neural network identification model.

[0020] The discharge event output module is used to obtain the spatial coordinates of the discharge source based on the identification results and sensor array data, and output the discharge type and positioning results to the monitoring terminal.

[0021] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0022] 1. By deploying UHF and high-frequency current sensors in the non-uniform electric field area of ​​GIS, combined with high sampling rate synchronous acquisition, high-fidelity capture of nanosecond-level weak discharge pulse signals is ensured; wavelet packet decomposition and entropy weight analysis are used to accurately identify key frequency bands with concentrated energy, effectively suppressing background noise; sparse modeling and reconstruction based on a self-built atomic dictionary and orthogonal matching pursuit algorithm significantly enhances weak discharge pulse signals; furthermore, short-time Fourier transform and wavelet packet energy mapping are integrated to construct a holographic time-frequency matrix spectrum, fully preserving the time-frequency coupling and local anomaly characteristics of partial discharge signals, improving the visualization clarity of signals and the accuracy of subsequent intelligent identification, possessing highly sensitive, robust, and high-resolution weak signal enhancement and visual diagnostic capabilities, suitable for intelligent partial discharge monitoring systems under complex operating conditions.

[0023] 2. By introducing a holographic time-frequency matrix map and a weak discharge neural network recognition model, accurate classification and spatial localization of weak discharge signals are achieved. This scheme integrates local preservation projection dimensionality reduction, temporal reconstruction, and dimensionality expansion techniques to construct a multidimensional tensor to preserve the time-frequency evolution characteristics of the signal. It utilizes deep learning modules such as multi-layer convolution, residual enhancement, and attention mechanisms to extract multi-scale coupling features of weak discharges, significantly improving the recognition capability of complex partial discharge maps with low signal-to-noise ratios. Combining a generalized cross-correlation algorithm and a nonlinear least squares localization method, sub-nanosecond time difference estimation and three-dimensional localization are achieved based on sensor array data. Finally, a structured result of discharge type and spatial coordinates is output, providing high-precision, interpretable intelligent monitoring and fault early warning support for power equipment. It possesses comprehensive advantages of high automation, high recognition accuracy, and small localization error, making it particularly suitable for online detection and maintenance decision-making of partial discharge events in complex operating environments. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the online detection method for GIS combined electrical appliances provided in an embodiment of this application.

[0025] Figure 2 This is a schematic diagram of the structure of the GIS combined electrical appliance online operation monitoring system provided in an embodiment of this application. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0027] Reference Figure 1 As shown, the present invention provides Figure 1 A schematic flowchart of the online monitoring method for GIS combined electrical appliances provided in this application embodiment includes the following steps:

[0028] S1. Acquire the original time-domain signal of partial discharge, perform signal enhancement processing on the original time-domain signal, extract the weak discharge pulse signal, and generate the enhanced discharge pulse signal.

[0029] Among them, UHF sensors and high-frequency current sensors are deployed in typical non-uniform electric field regions of GIS combined electrical appliances to synchronously collect the original time-domain signals of partial discharge during GIS operation; the sampling frequency of the original time-domain signals is not less than 1 GHz to ensure that they have time-domain resolution capability for nanosecond-level weak discharge pulses; the synchronously collected multi-channel original signals achieve timing consistency through a high-precision clock synchronization module, preserving the spatial distribution, phase information and relative timing characteristics of the signals.

[0030] The original time-domain signal is enhanced by extracting weak discharge pulse signals and generating enhanced discharge pulse signals, specifically as follows:

[0031] Multi-level wavelet packet decomposition is performed on the original time-domain signal to obtain multi-channel sub-band components;

[0032] Each sub-band component is sorted using the entropy weight method to identify the target frequency band signal component that highlights the pulse energy concentration;

[0033] Based on the temporal structure characteristics of the target frequency band components, an atomic dictionary is constructed;

[0034] Using the atomic dictionary as the basis for sparse modeling, the orthogonal matching pursuit algorithm is used to sparsely decompose the target frequency band signal components to obtain a set of sparse atoms and their corresponding sparse coefficients.

[0035] From the sparse atoms, sort them according to the matching value of each atom with the current signal residual;

[0036] The matching value is defined as the absolute value of the inner product of the atom and the current signal residual;

[0037] Atoms with matching values ​​greater than a preset threshold are selected as a subset of atoms with high matching degree with weak discharge pulse characteristics. Noise atoms with matching values ​​less than the preset threshold are removed. Based on the subset of atoms with high matching degree and its sparse coefficients, signal reconstruction is performed to generate a sparse enhanced signal.

[0038] Phase consistency recovery and time-domain reconstruction are performed on the sparse enhancement signal to output the enhanced discharge pulse signal.

[0039] It should be noted that an atom dictionary is a collection containing a large number of basic signal elements (called "atoms"), which are the basic building blocks for constructing signals. In partial discharge signal processing, the dictionary typically contains various pulse waveform templates that reflect the typical time-frequency characteristics of discharge pulses.

[0040] The Orthogonal Matching Pursuit (OMP) algorithm is a greedy sparse representation algorithm that aims to select a small number of atoms from a dictionary such that their linear combination best approximates the target signal. OMP makes the sparse representation more accurate and reduces residuals faster by orthogonally projecting the coefficients of all selected atoms in each iteration.

[0041] Sparse atoms refer to the subset of dictionary atoms selected for signal reconstruction during the OMP sparse algorithm. Compared to the entire dictionary, the number of sparse atoms is far less than the total number of atoms, reflecting the sparsity of the signal under the dictionary basis.

[0042] The sparsity coefficients are the weights of each sparse atom in a linear combination, reflecting the contribution of each atom to the signal representation.

[0043] Sparse atoms and their coefficient combinations constitute an approximate representation of the target signal, reflecting the key structure and characteristics of the signal. Weak discharge pulse signals are effectively extracted and represented through linear combinations of sparse atoms, achieving signal enhancement and noise reduction.

[0044] Furthermore, each sub-band component is sorted using the entropy weight method to identify the target frequency band signal component that highlights the pulse energy concentration, specifically:

[0045] The energy value of each sub-band component is calculated using the square of the L2 norm. And construct a sub-band energy vector, which is used to reflect the multi-scale frequency domain energy distribution characteristics of the discharge signal;

[0046] The energy value refers to the signal strength contained in the frequency band.

[0047] The sub-band energy vector is normalized to obtain the sub-band energy probability distribution, and the energy entropy value of each sub-band is calculated based on the information entropy.

[0048] The energy entropy value of each sub-band is converted into a weight. And combined with energy value Calculate the pulse energy concentration factor Construct a concentration vector;

[0049] The concentration vectors are sorted in descending order, and the sub-band components corresponding to the first few concentration vectors are selected as the target frequency band signal components.

[0050] Among them, the pulse energy concentration factor is a quantitative index that comprehensively reflects the energy magnitude and concentration of a certain sub-band signal in the multi-band analysis of partial discharge signals.

[0051] It should be noted that by fully utilizing the multi-level characteristics of wavelet packet decomposition to achieve joint time-frequency analysis, the adaptability to complex non-stationary signals is enhanced. Furthermore, by fusing energy values ​​and entropy information, the limitations of relying solely on energy thresholds are avoided, enabling more accurate capture of low-energy but highly bursty partial discharge signals and improving detection sensitivity. Moreover, by screening target frequency band components, noise bands are effectively eliminated, improving signal quality and laying a solid foundation for subsequent atomic dictionary construction and sparse enhancement analysis.

[0052] Furthermore, the orthogonal matching pursuit algorithm is used to perform sparse decomposition on the target frequency band signal components to obtain a set of sparse atoms and their corresponding sparse coefficients, specifically:

[0053] The target frequency band signal component is used as the initial residual. In each iteration, the atom corresponding to the largest absolute value of the inner product between the current residual and all atoms in the dictionary is selected, and the selected atom is added to the set of selected atoms.

[0054] After each iteration, based on the current set of selected atoms, the sparsity coefficients of the selected atoms are obtained by solving the least squares problem.

[0055] After the preset termination condition is met, a sparse subset of atoms consisting of a small number of atoms and their corresponding sparse coefficients are output for subsequent pulse enhancement and reconstruction processing steps, thereby realizing feature extraction and enhanced representation of the partial discharge pulse signal.

[0056] S2, a holographic time-frequency matrix spectrum is constructed based on the enhanced discharge pulse signal, specifically as follows:

[0057] The enhanced discharge pulse signal is windowed and then subjected to a short-time Fourier transform to obtain the initial time-frequency distribution matrix.

[0058] Among them, the windowing operation refers to using a sliding window function of finite length to segment and weight a continuous time signal or discrete time sequence along the time axis. The purpose is to divide the entire non-stationary signal into short-time segments that are approximately stationary, thus making it suitable for local spectrum analysis.

[0059] The enhanced discharge pulse signal is subjected to discrete wavelet packet transform, and the signal is decomposed layer by layer to a preset number of wavelet packet decomposition layers to obtain several frequency band sub-signals;

[0060] Energy mapping is performed on each frequency band sub-signal to construct the wavelet packet frequency band energy matrix;

[0061] The initial time-frequency distribution matrix and wavelet packet frequency band energy matrix are spatially aligned, and feature fusion is performed to generate a holographic time-frequency matrix map, so as to achieve a clear visualization of weak features and disturbance signals in the time domain signal.

[0062] The initial time-frequency distribution matrix refers to the two-dimensional spectral structure obtained after performing a short-time Fourier transform (STFT) on the enhanced discharge pulse signal. It reflects the energy distribution changes of the discharge signal in both time and frequency dimensions. By performing a discrete wavelet packet transform, finer-grained frequency band energy changes can be captured, which is particularly suitable for identifying the spectral leakage characteristics corresponding to different types of discharges.

[0063] A holographic time-frequency matrix map refers to a matrix map formed by fusing multi-resolution frequency band information from wavelet packet decomposition (DWPT) with an initial STFT map, ultimately possessing both global and local features and complete time-frequency information. It preserves the coupling information of weak discharge signals in the time, frequency, and energy dimensions.

[0064] It should be noted that by jointly applying short-time Fourier transform and discrete wavelet packet decomposition to the enhanced discharge pulse signal, the synergistic extraction of global time-frequency distribution and local fine-grained features was achieved, constructing a high-resolution holographic time-frequency matrix map. Based on spatial alignment and feature fusion, this map, enhanced by a two-dimensional enhancement algorithm, further improves the visualization clarity and resolution of weak signals, effectively enhancing the presentation of weak anomalous features in partial discharge signals, and significantly improving the detectability of weak discharge events in complex backgrounds and the accuracy of subsequent identification models.

[0065] S3 uses the holographic time-frequency matrix spectrum as input and identifies the type of weak discharge based on the weak discharge neural network recognition model, specifically:

[0066] The holographic time-frequency matrix spectrum is subjected to dimensionality reduction using a local preservation projection algorithm to extract low-dimensional feature representations with high information entropy;

[0067] The low-dimensional feature representation is temporally reconstructed and its dimensions expanded to construct a tensor containing temporal correlation and spectral structure. This tensor is used to adapt to the input interface of the subsequent weak discharge neural network, thereby achieving deep fusion of multi-dimensional time-frequency features.

[0068] Temporal reconstruction refers to reorganizing the feature vector sequence after dimensionality reduction into a time step sequence matrix with time dimension according to its original sampling time order, so that each frame (or sliding window) corresponds to a time slice, thereby preserving the temporal evolution characteristics of weak discharge signals.

[0069] Dimension expansion refers to expanding a current low-dimensional feature sequence into a four-dimensional tensor form that adapts to the input structure of a neural network by adding redundant or mapped dimensions.

[0070] Based on the tensor as input, a weak discharge neural network identification model is constructed. The weak discharge neural network identification model includes a shallow convolutional feature extraction network, a deep residual feature extraction module, an attention mechanism module, and a fully connected neural network classification layer to extract the coupling features of weak discharge signals in the time-frequency space.

[0071] A dataset of labeled weak discharge sample maps is obtained to conduct supervised learning training on the model. During the training process, the model is jointly optimized by the cross-entropy loss function and the Adam optimizer. The dataset of weak discharge sample maps includes typical samples of different discharge types, including but not limited to floating potential discharge, surface discharge, and crack partial discharge.

[0072] The trained weak discharge neural network identification model analyzes the holographic time-frequency matrix spectrum to be identified and outputs the corresponding weak discharge type classification results, which include the probability value of each type.

[0073] The output probability value determines whether the weak discharge type can be identified. If the maximum probability value is lower than the preset threshold, manual auxiliary diagnosis is triggered. If it is higher than the threshold, the final identification type is output. The identification types include floating potential discharge, surface discharge, crack partial discharge and air gap discharge.

[0074] It should be noted that by introducing a local preservation projection algorithm to reduce the dimensionality of the high-dimensional spectrum and extracting low-dimensional feature representations with high information entropy, and combining temporal reconstruction and dimensional expansion strategies to construct a tensor input containing time-frequency coupling information, the temporal evolution characteristics and spectral structure features of weak discharge signals can be effectively preserved. Furthermore, deep learning modules such as convolutional feature extraction, residual enhancement, attention weighting, and fully connected classification are used to perform multi-level modeling of the input features, thereby improving the ability to identify complex weak discharge types.

[0075] Furthermore, the trained weak discharge neural network recognition model analyzes the holographic time-frequency matrix spectrum to be identified and outputs the corresponding weak discharge type classification results, specifically:

[0076] The tensor is input into a shallow convolutional feature extraction network, and edge texture, local frequency band enhancement and high energy region features are extracted through several convolutional layers and pooling layers to form the first stage of the partial discharge feature map tensor.

[0077] The partial discharge feature map tensor is input to the deep residual feature extraction module, which includes multiple residual units. Through a jump connection structure, it retains shallow spatial detail information, enhances the feature expression depth, and outputs a deep tensor containing time-frequency semantic features.

[0078] The deep tensor is input into the attention mechanism module, which includes a channel attention module and a spatial attention module, and outputs a weighted feature map tensor.

[0079] Among them, the attention mechanism module is used to assign weighted responses to different frequency band channels, and the spatial attention module is used to highlight the spatial salience of the discharge region in the spectrum;

[0080] The weighted feature map tensor is transformed into a one-dimensional feature vector by global average pooling and feature flattening.

[0081] The one-dimensional feature vector is input into the classification layer of a fully connected neural network, and the probability value vector of each weak discharge type is output. The classification layer is composed of multiple fully connected layers and activation functions.

[0082] The probability vector is normalized using the Softmax function so that the sum of the probabilities of each category is 1. Finally, the weak discharge type corresponding to the highest probability is output as the identification result.

[0083] It should be noted that by constructing a multi-layered, modular weak discharge neural network recognition structure, deep feature extraction and classification of holographic time-frequency matrix maps are performed, fully exploring the local texture, frequency band structure, and spatial saliency information of weak discharge signals in the time-frequency domain. Basic local features are extracted through shallow convolution, residual modules enhance deep semantic expression, and an attention mechanism dynamically focuses on key frequency bands and discharge regions. Finally, a fully connected classifier outputs high-confidence classification results. This scheme effectively improves the accuracy and robustness of weak discharge type recognition, and is particularly suitable for processing low signal-to-noise ratio and complex partial discharge map data. It achieves an efficient and interpretable intelligent recognition process from map to category, providing reliable technical support for intelligent monitoring and fault early warning of power system insulation status.

[0084] S4, based on the identification results and sensor array data, obtains the spatial coordinates of the discharge source and outputs the discharge type and location results to the monitoring terminal, specifically:

[0085] The discharge occurrence time is obtained based on the model's recognition results, and the sensor array data for the corresponding time period is retrieved. The sensor array data includes spatial layout coordinate information, discharge response waveforms of each channel, and their absolute timestamps.

[0086] Using timestamps, the arrival time difference between each channel is calculated based on a generalized cross-correlation algorithm to form a time difference matrix;

[0087] Based on the time difference matrix and the spatial layout coordinate information of the sensor array, a set of spatial positioning equations constrained by time difference is established, and the nonlinear least squares method is used to solve them to obtain the spatial coordinate position of the discharge source.

[0088] The discharge type identification results are fused with the spatial coordinates of the discharge source to form a complete structured result set of discharge events;

[0089] The structured result set is output to the power equipment condition monitoring terminal through the communication module for remote monitoring, alarm push and risk assessment.

[0090] Reference Figure 2 As shown, the present invention provides Figure 2 The schematic diagram of the GIS combined electrical appliance online monitoring system provided in this application embodiment includes a signal enhancement module, a holographic time-frequency matrix map construction module, a weak discharge type identification module, and a discharge event output module. The modules are interconnected.

[0091] The signal enhancement module is used to acquire the original time-domain signal of partial discharge, perform signal enhancement processing on the original time-domain signal, extract the weak discharge pulse signal, and generate the enhanced discharge pulse signal.

[0092] The holographic time-frequency matrix map construction module is used to construct a holographic time-frequency matrix map based on the enhanced discharge pulse signal;

[0093] The weak discharge type identification module is used to identify the weak discharge type by taking the holographic time-frequency matrix spectrum as input and based on the weak discharge neural network identification model.

[0094] The discharge event output module is used to obtain the spatial coordinates of the discharge source based on the identification results and sensor array data, and output the discharge type and positioning results to the monitoring terminal.

[0095] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0096] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0097] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0098] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0099] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for online detection of GIS combined electrical appliances, characterized in that, Includes the following steps: The original time-domain signal of partial discharge is acquired, and signal enhancement processing is performed on the original time-domain signal to extract the weak discharge pulse signal and generate the enhanced discharge pulse signal, specifically as follows: Multi-level wavelet packet decomposition is performed on the original time-domain signal to obtain multi-channel sub-band components; Each sub-band component is sorted using the entropy weight method to identify the target frequency band signal components; Based on the temporal structure characteristics of the target frequency band components, an atomic dictionary is constructed; Using the atomic dictionary as the basis for sparse modeling, the orthogonal matching pursuit algorithm is used to sparsely decompose the target frequency band signal components to obtain sparse atoms and their corresponding sparse coefficients. From the sparse atoms, sort them according to the matching value of each atom with the current signal residual; Atoms with matching values ​​greater than a preset threshold are selected as a subset of atoms with a high degree of matching with the characteristics of weak discharge pulses; Based on the highly matched subset of atoms and their sparse coefficients, signal reconstruction is performed to generate a sparse enhanced signal; Phase consistency recovery and time-domain reconstruction are performed on the sparse enhanced signal to output an enhanced discharge pulse signal; A holographic time-frequency matrix spectrum is constructed based on the enhanced discharge pulse signal, specifically as follows: The enhanced discharge pulse signal is windowed and then subjected to a short-time Fourier transform to obtain the initial time-frequency distribution matrix. The enhanced discharge pulse signal is subjected to discrete wavelet packet transform, and the signal is decomposed layer by layer to a preset number of wavelet packet decomposition layers to obtain several frequency band sub-signals; Energy mapping is performed on each frequency band sub-signal to construct the wavelet packet frequency band energy matrix; The initial time-frequency distribution matrix and wavelet packet frequency band energy matrix are spatially aligned, and feature fusion is performed to generate a holographic time-frequency matrix map; Using the holographic time-frequency matrix spectrum as input, the weak discharge type is identified based on the weak discharge neural network identification model. Based on the identification results and sensor array data, the spatial coordinates of the discharge source are obtained, and the discharge type and positioning results are output to the monitoring terminal.

2. The online detection method for GIS combined electrical appliances according to claim 1, characterized in that, The process of sorting each sub-frequency band component using the entropy weight method to identify the target frequency band signal component highlighting pulse energy concentration is specifically as follows: Calculate the energy value of each sub-band component and construct the sub-band energy vector; The sub-band energy vector is normalized to obtain the sub-band energy probability distribution, and the energy entropy value of each sub-band is calculated based on the information entropy. The energy entropy value of each sub-band is converted into a weight, and the pulse energy concentration factor is calculated by combining the energy value to construct a concentration vector; The concentration vectors are sorted in descending order, and the sub-band components corresponding to the first few concentration vectors are selected as the target frequency band signal components.

3. The online detection method for GIS combined electrical appliances according to claim 2, characterized in that, The orthogonal matching pursuit algorithm is used to perform sparse decomposition on the target frequency band signal components to obtain sparse atoms and their corresponding sparse coefficients. Specifically: The target frequency band signal component is used as the initial residual. In each iteration, the atom corresponding to the largest absolute value of the inner product between the current residual and all atoms in the dictionary is selected, and the selected atom is added to the set of selected atoms. After each iteration, based on the current set of selected atoms, the sparsity coefficients corresponding to the selected atoms are obtained by the least squares method. After the preset termination condition is met, the output is a sparse subset of atoms consisting of a small number of atoms and their corresponding sparse coefficients.

4. The online detection method for GIS combined electrical appliances according to claim 1, characterized in that, The process of using a holographic time-frequency matrix as input and identifying weak discharge types based on a weak discharge neural network identification model specifically involves: The holographic time-frequency matrix spectrum is reduced in dimensionality using the local preservation projection method to extract low-dimensional feature representations with high information entropy; The low-dimensional feature representation is temporally reconstructed and dimensionally expanded to construct a tensor containing temporal correlation and spectral structure. Based on the tensor as input, a weak discharge neural network recognition model is constructed. The weak discharge neural network recognition model includes a shallow convolutional feature extraction network, a deep residual feature extraction module, an attention mechanism module, and a fully connected neural network classification layer. We obtained a dataset of labeled weak discharge sample maps to train the model using supervised learning. The trained weak discharge neural network identification model analyzes the holographic time-frequency matrix spectrum to be identified and outputs the corresponding weak discharge type classification results, which include the probability value of each type. The probability value output determines whether the weak discharge type can be identified. If it is higher than the threshold, the final identification type is output.

5. The online detection method for GIS combined electrical appliances according to claim 4, characterized in that, The trained weak discharge neural network identification model analyzes the holographic time-frequency matrix spectrum to be identified and outputs the corresponding weak discharge type classification results, specifically: The tensor is input into a shallow convolutional feature extraction network, and features are extracted through several convolutional and pooling layers to form the first-stage partial discharge feature map tensor. The partial discharge feature map tensor is input into the deep residual feature extraction module, and the output is a deep tensor containing time-frequency semantic features. The deep tensor is input into the attention mechanism module, and the weighted feature map tensor is output. The weighted feature map tensor is transformed into a one-dimensional feature vector by global average pooling and feature flattening. The one-dimensional feature vector is input into the classification layer of a fully connected neural network, and the probability value vector of each weak discharge type is output. The probability vector is normalized using the Softmax function so that the sum of the probabilities of each category is 1. Finally, the weak discharge type corresponding to the highest probability is output as the identification result.

6. The online detection method for GIS combined electrical appliances according to claim 1, characterized in that, The process of obtaining the spatial coordinates of the discharge source based on the identification results and sensor array data, and outputting the discharge type and positioning results to the monitoring terminal, specifically involves: The discharge occurrence time is obtained based on the model's recognition results, and the sensor array data for the corresponding time period is retrieved. The sensor array data includes spatial layout coordinate information, discharge response waveforms of each channel, and their absolute timestamps. Using timestamps, the arrival time difference between each channel is calculated based on a generalized cross-correlation algorithm to form a time difference matrix; Based on the time difference matrix and combined with the spatial layout coordinate information of the sensor array, the spatial coordinate position of the discharge source is obtained by nonlinear least squares method. The discharge type identification results are fused with the spatial coordinates of the discharge source to form a structured result set of discharge events; The structured result set is output to the power equipment condition monitoring terminal through the communication module.

7. A system using the online detection method for GIS combined electrical appliances as described in any one of claims 1-6, characterized in that, It includes a signal enhancement module, a holographic time-frequency matrix map construction module, a weak discharge type identification module, and a discharge event output module, and the modules are interconnected: The signal enhancement module is used to acquire the original time-domain signal of partial discharge, perform signal enhancement processing on the original time-domain signal, extract the weak discharge pulse signal, and generate the enhanced discharge pulse signal. The holographic time-frequency matrix map construction module is used to construct a holographic time-frequency matrix map based on the enhanced discharge pulse signal; The weak discharge type identification module is used to identify the weak discharge type by taking the holographic time-frequency matrix spectrum as input and based on the weak discharge neural network identification model. The discharge event output module is used to obtain the spatial coordinates of the discharge source based on the identification results and sensor array data, and output the discharge type and positioning results to the monitoring terminal.

Images