Partial discharge location method and system based on radio waves

Abstract

The present invention relates to the field of data processing technology, and discloses a local discharge positioning method and system based on radio waves, which are used to improve the accuracy of local discharge positioning based on radio waves. The method includes: performing cluster analysis on local discharge feature vectors to obtain discharge event clustering results, and performing spatiotemporal correlation analysis on the discharge event clustering results to obtain discharge source candidate positions; arranging multiple directional antenna arrays in the monitoring area, collecting electromagnetic wave direction data in real time, obtaining arrival angle information, and performing triangulation positioning calculation on the arrival angle information to obtain preliminary spatial coordinates; performing error analysis on the preliminary spatial coordinates to obtain positioning accuracy evaluation data, and performing matching analysis on the positioning accuracy evaluation data and the discharge source candidate positions to obtain an optimized discharge source position; performing time series analysis on the optimized discharge source position to obtain discharge mode data, and performing intelligent recognition on the discharge mode data to obtain discharge type and severity evaluation results.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a local discharge locating method and system based on radio waves. Background Art

[0002] Partial discharge positioning technology plays an important role in power system equipment monitoring and fault diagnosis. Traditional partial discharge positioning methods mainly include acoustic positioning, ultra-high frequency positioning and optical positioning. The acoustic positioning method uses the sound wave signal generated by the discharge for positioning, but is greatly affected by environmental noise; the ultra-high frequency positioning method locates by capturing the electromagnetic waves generated by the discharge, and has high sensitivity; the optical positioning method uses the light signal generated by the discharge for positioning, but the application scenarios are limited. In recent years, with the development of wireless communication technology, partial discharge positioning methods based on radio waves have gradually attracted attention. This method arranges multiple receiving antennas and uses information such as arrival time difference or arrival angle for positioning calculation.

[0003] However, the existing radio wave-based local discharge positioning methods still have some shortcomings. First, in a complex electromagnetic environment, the received signals often contain a lot of noise and interference, which affects the positioning accuracy. Secondly, traditional signal processing and feature extraction methods are difficult to fully utilize the time-frequency characteristics of local discharge signals, resulting in incomplete feature expression. Furthermore, the existing positioning algorithms are mostly based on a single principle and are difficult to adapt to positioning needs in different scenarios. In addition, for the type identification and severity assessment of the discharge source, the existing methods often rely on expert experience and lack intelligent and automated means. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a method and system for locating partial discharge based on radio waves, so as to improve the accuracy of locating partial discharge based on radio waves.

[0005] The present invention provides a local discharge positioning method based on radio waves, comprising: collecting and processing multi-band broadband signals to obtain original electromagnetic signal data, preprocessing and denoising the original electromagnetic signal data to obtain cleaned signal data; performing time-frequency analysis on the cleaned signal data to obtain time-frequency feature data, and extracting features from the time-frequency feature data to obtain local discharge feature vectors; performing cluster analysis on the local discharge feature vectors to obtain discharge event clustering results, and performing time-space correlation analysis on the discharge event clustering results to obtain discharge source candidate positions; arranging multiple directional antenna arrays in a monitoring area, collecting electromagnetic wave direction data in real time to obtain arrival angle information, and performing triangulation positioning calculation on the arrival angle information to obtain preliminary spatial coordinates; performing error analysis on the preliminary spatial coordinates to obtain positioning accuracy evaluation data, and performing matching analysis on the positioning accuracy evaluation data and the discharge source candidate positions to obtain optimized discharge source positions; performing time series analysis on the optimized discharge source positions to obtain discharge mode data, and performing intelligent recognition on the discharge mode data to obtain discharge type and severity evaluation results.

[0006] The present invention also provides a local discharge locating system based on radio waves, comprising:

[0007] An acquisition module is used to acquire and process multi-band broadband signals to obtain raw electromagnetic signal data, and to pre-process and reduce noise on the raw electromagnetic signal data to obtain cleaned signal data;

[0008] An analysis module, used for performing time-frequency analysis on the cleaning signal data to obtain time-frequency feature data, and performing feature extraction on the time-frequency feature data to obtain a local discharge feature vector;

[0009] A clustering module, used for performing cluster analysis on the local discharge feature vectors to obtain a discharge event clustering result, and performing spatiotemporal correlation analysis on the discharge event clustering result to obtain a discharge source candidate location;

[0010] A calculation module is used to arrange multiple directional antenna arrays in the monitoring area, collect electromagnetic wave direction data in real time, obtain arrival angle information, and perform triangulation positioning calculation on the arrival angle information to obtain preliminary spatial coordinates;

[0011] A matching module, used for performing error analysis on the preliminary spatial coordinates to obtain positioning accuracy evaluation data, and performing matching analysis on the positioning accuracy evaluation data and the candidate position of the discharge source to obtain an optimized discharge source position;

[0012] The identification module is used to perform a time series analysis on the optimized discharge source position to obtain discharge mode data, and to perform intelligent identification on the discharge mode data to obtain a discharge type and severity evaluation result.

[0013] In the technical solution provided by the present invention, the quality of the original electromagnetic signal is effectively improved by collecting and processing the multi-band broadband signal and pre-processing the noise reduction, and the cleaning signal data is subjected to time-frequency analysis and feature extraction, which can not only fully capture the time-frequency characteristics of the local discharge signal, but also effectively compress the data volume and improve the processing efficiency. By performing cluster analysis and spatiotemporal correlation analysis on the local discharge feature vector, different types of discharge events can be effectively identified and classified, and the possible location of the discharge source can be preliminarily determined, which greatly narrows the search range. Multiple directional antenna arrays are arranged in the monitoring area to collect electromagnetic wave direction data in real time. Combined with triangulation positioning calculation, the preliminary spatial coordinates of the discharge source can be quickly and accurately determined. Error analysis and matching analysis are performed on the preliminary spatial coordinates, which can not only evaluate the reliability of the positioning result, but also further optimize the positioning result by comparing with the candidate location of the discharge source, significantly improving the positioning accuracy. Finally, the timing analysis and intelligent identification of the optimized discharge source position can not only extract the discharge mode data, but also accurately judge the discharge type and severity. Through multi-step data processing and analysis, the whole process automation from signal acquisition to discharge source positioning to discharge type identification is realized, which greatly improves the efficiency and accuracy of local discharge detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0015] Figure 1 The figure is a flow chart of a method for locating partial discharge based on radio waves in an embodiment of the present invention.

[0016] Figure 2 Schematic diagram of a local discharge locating system based on radio waves in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0019] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0020] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , Figure 1 is a flow chart of a method for locating partial discharge based on radio waves according to an embodiment of the present invention, such as Figure 1 As shown, the following steps are included:

[0021] S101, collecting and processing multi-band broadband signals to obtain original electromagnetic signal data, and preprocessing and denoising the original electromagnetic signal data to obtain cleaned signal data;

[0022] S102, performing time-frequency analysis on the cleaning signal data to obtain time-frequency feature data, and performing feature extraction on the time-frequency feature data to obtain a partial discharge feature vector;

[0023] S103, performing cluster analysis on the local discharge feature vectors to obtain a discharge event clustering result, and performing spatiotemporal correlation analysis on the discharge event clustering result to obtain a discharge source candidate location;

[0024] S104, arranging multiple directional antenna arrays in the monitoring area, collecting electromagnetic wave direction data in real time, obtaining arrival angle information, and performing triangulation positioning calculation on the arrival angle information to obtain preliminary spatial coordinates;

[0025] S105, performing error analysis on the preliminary spatial coordinates to obtain positioning accuracy evaluation data, and performing matching analysis on the positioning accuracy evaluation data and the candidate positions of the discharge source to obtain the optimized discharge source position;

[0026] S106, performing a time series analysis on the optimized discharge source position to obtain discharge pattern data, and performing intelligent recognition on the discharge pattern data to obtain a discharge type and severity assessment result.

[0027] It should be noted that during the acquisition process, a broadband antenna array is used to simultaneously capture electromagnetic signals in the range of 1MHz-3GHz, and the receiving sensitivity is adjusted in real time through adaptive gain control technology to ensure that the best signal-to-noise ratio can be obtained in different environments. Subsequently, the original electromagnetic signal data is preprocessed and denoised, including bandpass filtering, wavelet transform denoising and empirical mode decomposition, and finally the cleaned signal data is obtained. The cleaned signal data is subjected to time-frequency analysis, and the time-frequency spectrum is generated by short-time Fourier transform, and the energy density distribution is calculated. The main frequency band intervals are determined by peak detection and cluster analysis, and then these intervals are time series sliced ​​and wavelet packet decomposed. The entropy value of the wavelet packet coefficients is calculated to obtain the information entropy feature, and the dimension reduction is performed through principal component analysis. Finally, the reduced-dimensional features are subjected to nonlinear mapping and feature selection, and the local discharge feature vector is obtained by fusion.

[0028] The local discharge feature vectors are clustered and the initial clustering results are obtained using the density peak clustering algorithm. The clustering quality is evaluated by the silhouette coefficient, and the clustering parameters are adaptively adjusted according to the evaluation results. The clustering is re-executed to obtain the clustering results of the discharge events. Subsequently, the clustering results are analyzed by timestamp and spatial distribution to obtain the discharge cycle pattern and the coordinates of the high-incidence area, respectively. Based on this information, a spatiotemporal association network is constructed, and the spatiotemporal clustering clusters are obtained by the spectral clustering algorithm. Finally, the cluster center coordinates are calculated and spatial interpolation is performed to obtain the candidate location of the discharge source. Multiple directional antenna arrays are arranged in the monitoring area to collect electromagnetic wave direction data in real time. First, the monitoring area is gridded, the antenna layout scheme is determined, and the antenna array network is calibrated. Electromagnetic wave signals are collected by the antenna array, and phase contrast is performed on the multi-channel raw data. The direction of arrival is estimated in combination with the calibration parameters to obtain the arrival angle information. The arrival angle information is smoothed, and then the candidate spatial point set is obtained by multi-station intersection calculation. The candidate points are fitted using the least squares method to obtain the optimal fitting points, and the coordinates are transformed to obtain the preliminary spatial coordinates.

[0029] The error analysis of the preliminary spatial coordinates is carried out, and the coordinate distribution samples are generated by the Monte Carlo simulation method. The error ellipse parameters are obtained by statistical analysis. The positioning accuracy index is calculated according to the error ellipse parameters to obtain the accuracy assessment matrix. The accuracy assessment matrix is ​​segmented by threshold to determine the credible area boundary. The spatial relationship between the credible area boundary and the candidate location of the discharge source is analyzed to obtain the overlapping area, and the density of the overlapping area is estimated. The most likely discharge source location is determined by maximum likelihood estimation, and a local search is performed to obtain a refined location estimate. Multi-source information fusion is performed on the refined location estimate to obtain a comprehensive score, and the scores are sorted and screened, and finally the optimized discharge source location is obtained by weighted average. Finally, the optimized discharge source location is analyzed in time series. First, the time window is segmented to obtain time series data fragments. These fragments are Fourier transformed to obtain frequency domain feature sequences. The characteristic frequency set is obtained by peak extraction, and the discharge frequency template is obtained by pattern matching. The time domain reconstruction is performed according to the frequency template to obtain a typical discharge waveform, and the waveform is parameterized to obtain a discharge feature parameter set. The feature parameter set is analyzed in time series to obtain discharge pattern data. Cluster analysis was performed on the discharge pattern data to obtain the discharge pattern categories. A decision tree model was built based on the discharge pattern categories to obtain the preliminary judgment results of the discharge type, and the discharge type probability distribution was obtained through fuzzy reasoning. Finally, the discharge type probability distribution and the discharge characteristic parameter set were comprehensively evaluated to obtain the discharge type and severity evaluation results.

[0030] For example, this method is used to locate partial discharge in the GIS equipment of a high-voltage substation. First, 8 broadband antennas are arranged around the GIS to collect electromagnetic signals in the range of 1MHz-3GHz. Through adaptive sampling, the original signal sampling rate is 10GS / s and the data volume is 100MB / s. After bandpass filtering and wavelet denoising, the signal data volume is reduced to 10MB / s. The cleaned signal is short-time Fourier transformed to generate a time-frequency spectrum, and the energy is mainly concentrated in the 300MHz-500MHz frequency band. Through wavelet packet decomposition and entropy value calculation, a 128-dimensional feature vector is obtained. The feature vector is clustered using the density peak clustering algorithm. The initial number of cluster centers is 5, and after adaptive adjustment, it is determined to be 3 clusters. The spatiotemporal correlation analysis shows that the discharge event is periodic with a period of about 10ms, and is mainly concentrated in three areas of the GIS in space. The triangulated positioning calculation is performed through the arrival angle information of multiple directional antenna arrays. The error ellipse of the preliminary positioning result has a major axis of 0.5m and a minor axis of 0.3m. After matching analysis with candidate locations, the accuracy of the optimized discharge source location was improved to ±0.1m. Finally, through timing analysis and pattern recognition, the discharge type was identified as suspended potential discharge, with a moderate severity level, which required close monitoring.

[0031] By executing the above steps, the quality of the original electromagnetic signal is effectively improved by collecting and processing the multi-band broadband signal and pre-processing the noise reduction. The time-frequency analysis and feature extraction of the cleaning signal data can not only fully capture the time-frequency characteristics of the partial discharge signal, but also effectively compress the data volume and improve the processing efficiency. By clustering and spatiotemporal correlation analysis of the partial discharge feature vector, different types of discharge events can be effectively identified and classified, and the possible location of the discharge source can be preliminarily determined, which greatly narrows the search range. Multiple directional antenna arrays are arranged in the monitoring area to collect electromagnetic wave direction data in real time. Combined with triangulation positioning calculation, the preliminary spatial coordinates of the discharge source can be quickly and accurately determined. Error analysis and matching analysis of the preliminary spatial coordinates can not only evaluate the reliability of the positioning results, but also further optimize the positioning results by comparing with the candidate positions of the discharge source, significantly improving the positioning accuracy. Finally, the timing analysis and intelligent identification of the optimized discharge source position can not only extract the discharge mode data, but also accurately judge the discharge type and severity. Through multi-step data processing and analysis, the whole process from signal acquisition to discharge source positioning to discharge type identification is realized, which greatly improves the efficiency and accuracy of partial discharge detection.

[0032] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0033] (1) Adaptively sample the multi-band broadband signal to obtain the original sampling data, and perform spectrum analysis on the original sampling data to obtain the frequency distribution characteristics;

[0034] (2) Band-pass filtering is performed on the frequency distribution characteristics to obtain the target frequency band signal, and the target frequency band signal is reconstructed in the time domain to obtain the time domain reconstructed signal;

[0035] (3) Performing wavelet transform on the time domain reconstructed signal to obtain wavelet coefficients, and performing threshold denoising on the wavelet coefficients to obtain denoised wavelet coefficients;

[0036] (4) Reconstruct the denoised wavelet coefficients by inverse wavelet transform to obtain a preliminary denoised signal, and perform empirical mode decomposition on the preliminary denoised signal to obtain an eigenmode function set;

[0037] (5) Perform correlation analysis on the eigenmode function set to obtain the main modal components, and reconstruct and superimpose the main modal components to obtain the reconstructed signal;

[0038] (6) Perform mean filtering on the reconstructed signal to obtain a smoothed signal, and normalize the smoothed signal to obtain cleaned signal data.

[0039] Specifically, during the adaptive sampling process, the sampling rate is dynamically adjusted according to the sparse characteristics of the signal to ensure that the amount of data is reduced while retaining key information. The raw sampling data obtained by sampling is then subjected to spectral analysis, and the time domain signal is converted into the frequency domain through the fast Fourier transform (FFT) to obtain the frequency distribution characteristics. The spectrum analysis results show the energy distribution of the signal at different frequencies, which helps to identify the main frequency bands of the partial discharge signal. The frequency distribution characteristics are band-pass filtered to filter out irrelevant frequency bands and retain the target frequency band signal. The center frequency and bandwidth of the band-pass filter are designed according to the typical frequency characteristics of the partial discharge signal, usually covering a range of several hundred kHz to several MHz. The filtered signal is reconstructed in the time domain by inverse Fourier transform to obtain a time domain reconstructed signal, which effectively removes noise and interference outside the frequency band.

[0040] The time-domain reconstructed signal is subjected to wavelet transform, and a wavelet basis function suitable for the characteristics of the partial discharge signal, such as Daubechies wavelet or Morlet wavelet, is selected to decompose the signal into wavelet coefficients of different scales. Wavelet transform can effectively capture the local time-frequency characteristics of the signal, and is particularly suitable for analyzing non-stationary signals such as partial discharge. After obtaining the wavelet coefficients, the wavelet coefficients are processed by threshold denoising method. The selection of thresholds is usually based on the statistical characteristics of the wavelet coefficients, such as the use of the general threshold method or the SURE threshold method to effectively remove the noise components in the wavelet domain and obtain the denoised wavelet coefficients. Subsequently, the denoised wavelet coefficients are reconstructed by inverse wavelet transform to obtain a preliminary denoised signal. The denoised wavelet coefficients are recombined and restored to the time-domain signal, while retaining the important features of the signal and suppressing the noise. The preliminary denoised signal is subjected to empirical mode decomposition (EMD), which is an adaptive signal processing method that is particularly suitable for processing nonlinear and non-stationary signals. EMD decomposes the signal into a series of intrinsic mode functions (IMFs) and a residual term, and each IMF represents an oscillation mode of different scales in the signal.

[0041] The obtained intrinsic mode function set is subjected to correlation analysis, the correlation coefficient between each IMF and the original signal is calculated, and the main modal components most related to the local discharge characteristics are identified. Correlation analysis helps to distinguish the effective components from the noise components in the signal. The IMF with high correlation is selected as the main modal component, and the reconstructed signal is obtained by reconstructing and superimposing these components. The key information in the signal is effectively retained, while the noise and interference are further suppressed. Finally, the reconstructed signal is mean filtered, and the local average value is calculated through a sliding window to obtain a smoothed signal. Mean filtering can effectively suppress high-frequency noise and make the signal profile smoother. The smoothed signal is normalized to scale the signal amplitude to a uniform range, usually [0,1] or [-1,1], to obtain the final cleaned signal data. Normalization processing helps the subsequent feature extraction and pattern recognition process.

[0042] For example, this method is applied to the GIS equipment of a 500kV substation for partial discharge signal processing. First, a broadband antenna is used to collect electromagnetic signals in the range of 1MHz-3GHz. Through adaptive sampling, the initial sampling rate is set to 10GS / s. Spectral analysis shows that the main energy is concentrated in the range of 300MHz-700MHz. Based on this, a bandpass filter is designed with a center frequency of 500MHz and a bandwidth of 400MHz. The db4 wavelet is used for wavelet transform and 5-layer decomposition is performed. The soft threshold method is used for threshold denoising, and the threshold is set to 3 times the standard deviation of the wavelet coefficient. EMD decomposition obtains 10 IMFs. Through correlation analysis, the first 3 IMFs with correlation coefficients greater than 0.6 are selected for reconstruction. The mean filter window size is set to 21 sampling points. After this series of processing, the background noise in the original signal is significantly reduced, the signal-to-noise ratio is increased from the initial 5dB to 15dB, and the time domain waveform of the partial discharge pulse is clearer, which is conducive to subsequent feature extraction and positioning analysis.

[0043] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0044] (1) Perform short-time Fourier transform on the cleaning signal data to obtain a time-frequency spectrum, and calculate the energy density of the time-frequency spectrum to obtain an energy distribution matrix;

[0045] (2) Perform peak detection on the energy distribution matrix to obtain a characteristic peak point set, and perform cluster analysis on the characteristic peak point set to obtain the main frequency band interval;

[0046] (3) Slice the time series of the main frequency band interval to obtain a time-frequency slice set, and perform wavelet packet decomposition on the time-frequency slice set to obtain wavelet packet coefficients;

[0047] (4) Calculate the entropy value of the wavelet packet coefficients to obtain the information entropy characteristics, and perform statistical analysis on the information entropy characteristics to obtain the entropy distribution characteristics;

[0048] (5) The entropy distribution features are reduced in dimension through the principal component analysis algorithm to obtain low-dimensional feature representation, and the low-dimensional feature representation is nonlinearly mapped to obtain high-order features;

[0049] (6) Perform feature selection on high-order features to obtain key feature subsets, and perform feature fusion on the key feature subsets to obtain the local discharge feature vector.

[0050] It should be noted that the short-time Fourier transform calculates the local spectrum at each time point by sliding the window function on the time domain signal to obtain a time-frequency spectrum. This method can simultaneously reflect the characteristics of the signal in the time and frequency dimensions, and is particularly suitable for analyzing non-stationary signals such as partial discharge. The energy density of the obtained time-frequency spectrum is calculated, and the energy value of each time-frequency point is calculated to form an energy distribution matrix. The energy distribution matrix intuitively shows the distribution of signal energy on the time-frequency plane, which helps to identify the time-frequency characteristics of partial discharge events. Peak detection is performed on the energy distribution matrix to identify the time-frequency points where the energy is concentrated. Peak detection usually uses a local maximum search algorithm and sets an appropriate threshold to exclude weak interference. The obtained characteristic peak point set represents a significant event in the signal. Cluster analysis is performed on these characteristic peak point sets, such as using the K-means or DBSCAN algorithm to classify similar peak points to obtain the main frequency band intervals. These frequency band intervals reflect the main frequency components of the partial discharge signal.

[0051] The time series of the main frequency band intervals are sliced, and the time-frequency plane is divided into a series of time-frequency blocks to obtain a set of time-frequency slices. Each slice contains the signal information of a specific frequency band in a specific time period. Wavelet packet decomposition is performed on these time-frequency slice sets. Wavelet packet decomposition is an extension of wavelet transform and provides a more detailed frequency division. By selecting appropriate wavelet basis functions and decomposition layers, a series of wavelet packet coefficients are obtained. Wavelet packet decomposition can more finely characterize the characteristics of the signal at different time-frequency scales. Entropy values ​​are calculated for wavelet packet coefficients. Entropy values ​​are an important indicator for describing signal complexity and uncertainty. Commonly used entropies include Shannon entropy and energy entropy. Entropy value calculation reflects the energy distribution of the signal in each sub-band, and information entropy features are obtained. Statistical analysis is performed on the information entropy features, such as calculating statistical quantities such as mean, variance, skewness, and kurtosis to obtain entropy value distribution characteristics. These features can describe the complexity and energy distribution pattern of partial discharge signals from a statistical perspective.

[0052] The principal component analysis algorithm is used to reduce the dimension of the entropy distribution features. Principal component analysis converts high-dimensional features that may be correlated into linearly independent low-dimensional features through orthogonal transformation, retaining the main information of the data. The resulting low-dimensional feature representation reduces data redundancy and highlights key information. Nonlinear mapping of the low-dimensional feature representation, such as using kernel methods or neural networks, can capture the nonlinear relationship between features and obtain high-order features. Nonlinear mapping can discover complex patterns hidden in the original features.

[0053] Finally, feature selection is performed on high-order features, such as using recursive feature elimination or correlation-based feature selection methods to screen out the most discriminative feature subsets. Feature selection helps to remove redundant and irrelevant features and improve the efficiency and accuracy of subsequent processing. Feature fusion is performed on the selected key feature subsets, such as using weighted summation or Fisher discriminant analysis to combine multiple features into a comprehensive feature vector. The final partial discharge feature vector contains both the time-frequency characteristics of the signal and reflects the energy distribution and complexity information, providing strong support for subsequent discharge positioning and identification.

[0054] For example, this method is applied to extract partial discharge features in the GIS equipment of a 330kV substation. First, the collected cleaning signal data is subjected to short-time Fourier transform, the window length is set to 1024 points, and the overlap rate is 50%. The obtained time-frequency spectrum shows that the main energy is concentrated in the 300MHz-700MHz frequency band. The energy density calculation uses a logarithmic scale to highlight the weak signal characteristics. The peak detection threshold is set to 3 times the average energy, and 57 characteristic peak points are detected. The DBSCAN algorithm is used for clustering, the parameter ε is set to 5MHz, and MinPts is set to 3, and 4 main frequency bands are obtained: 350-400MHz, 450-500MHz, 550-600MHz and 650-700MHz. These frequency bands are sliced ​​in time series, and the length of each slice is 100μs. The db4 wavelet basis function is used to perform a 4-layer wavelet packet decomposition to obtain the wavelet packet coefficients of 16 subbands. The Shannon entropy of each subband is calculated to obtain a 16-dimensional information entropy feature. Statistical analysis showed that the mean of the entropy value was 2.7 and the standard deviation was 0.5. The principal component analysis reduced the 16-dimensional features to 8 dimensions, retaining 95% of the variance information. Nonlinear mapping was performed through the radial basis function kernel to obtain 20-dimensional high-order features. Feature selection retained the 10 most discriminative features, and finally these 10 features were fused into a 5-dimensional local discharge feature vector through Fisher discriminant analysis. This feature vector not only reflects the main frequency band information of the signal, but also contains energy distribution and complexity characteristics, providing an effective feature representation for subsequent discharge positioning.

[0055] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0056] (1) Perform density peak clustering on the local discharge feature vectors to obtain the initial clustering results, and perform silhouette coefficient evaluation on the initial clustering results to obtain the clustering quality index;

[0057] (2) Adaptively adjust the clustering parameters according to the clustering quality index to obtain the optimized clustering parameters, and re-execute the density peak clustering by optimizing the clustering parameters to obtain the discharge event clustering results;

[0058] (3) Perform timestamp analysis on the clustering results of discharge events to obtain time series features, and perform periodicity detection on the time series features to obtain the discharge cycle pattern;

[0059] (4) Perform spatial distribution analysis on the clustering results of discharge events to obtain a spatial density map, and identify hot spots on the spatial density map to obtain the coordinates of high-incidence areas;

[0060] (5) Construct a spatiotemporal correlation network based on the discharge cycle pattern and the coordinates of the high-incidence area to obtain the correlation weight matrix, and perform spectral clustering on the correlation weight matrix to obtain the spatiotemporal clustering clusters;

[0061] (6) Calculate the centroid of the spatiotemporal clusters to obtain the cluster center coordinate set, and perform spatial interpolation on the cluster center coordinate set to obtain the candidate location of the discharge source.

[0062] Specifically, the density peak clustering algorithm first calculates the local density of each data point and the distance to the high density point . Local density The calculation formula is:

[0063]

[0064] in, It is a data point and The distance between is the cutoff distance.

[0065] distance The calculation formula is:

[0066]

[0067] For the point with the largest density, define .

[0068] By calculation ,choose γ The point with the larger value is taken as the cluster center. After obtaining the initial clustering results, the silhouette coefficient is used to evaluate the clustering quality. Silhouette coefficient The calculation formula is:

[0069]

[0070] in, Yes The average distance to other points in the same cluster, Yes The minimum value of the average distance to all points in the nearest other clusters. According to the clustering quality index, the clustering parameters are adaptively adjusted. The main adjustment parameters include the cutoff distance and the cluster center selection threshold. Through iterative optimization, the best clustering parameters are obtained, and the density peak clustering is re-executed to obtain the final discharge event clustering result.

[0071] The time stamp analysis of the discharge event clustering results is performed to extract the time series features of each cluster. The time series features include the time interval and frequency of the event. These features are periodically detected, and the discharge cycle pattern is identified by combining the autocorrelation function and the fast Fourier transform.

[0072] At the same time, the spatial distribution analysis of the clustering results of the discharge events is performed. The spatial density map is generated using the kernel density estimation method. The general form of the kernel density estimation function f(x) is:

[0073]

[0074] in, is the kernel function, is the bandwidth parameter, is the sample point. By analyzing the spatial density map, high-density areas are identified and the coordinates of high-incidence areas are obtained. Based on the discharge cycle pattern and the coordinates of high-incidence areas, a spatiotemporal association network is constructed. The nodes in the network represent discharge events, and the weights of the edges reflect the spatiotemporal correlation between events. After obtaining the association weight matrix, the spectral clustering algorithm is applied for spatiotemporal clustering. The core of spectral clustering is to calculate the eigenvectors of the Laplace matrix and cluster them in a low-dimensional space.

[0075] Finally, the centroid of the spatiotemporal clusters is calculated to obtain a cluster center coordinate set, and the cluster center coordinate set is spatially interpolated, such as using the Kriging interpolation method, to obtain a continuous distribution of candidate discharge source locations.

[0076] For example, this method is applied to localize partial discharge in a GIS device of a 500kV substation. First, the extracted 5-dimensional partial discharge feature vector is clustered by density peak. The initial cutoff distance is set to 2% of the average distance in the feature space, and the ρ and δ values ​​of each point are calculated. γThe top three points with the largest values ​​were taken as cluster centers, and three initial clusters were obtained. The silhouette coefficient was calculated, and the average value was 0.68. By adjusting the initial cutoff distance value, it was finally optimized to 0.015. After re-clustering, four discharge event clusters were obtained, and the average silhouette coefficient was increased to 0.75. Timestamp analysis showed that the discharge period of the first cluster was 10ms, which was related to the grid frequency; the second cluster showed a random distribution; the third and fourth clusters showed a periodicity of 30s. The spatial distribution analysis used a Gaussian kernel function, and the bandwidth h was set to 0.5m. Three high-incidence areas were identified, with coordinates of (2.5, 1.8, 1.2)m, (3.7, 2.2, 1.5)m, and (4.8, 2.5, 1.8)m.

[0077] When constructing a spatiotemporal association network, the association weight The calculation considers two dimensions: time and space:

[0078]

[0079] in, and is the time when the event occurred, and are the event space coordinates, and are the time and space scale parameters. Set 5ms, Spectral clustering obtains five spatiotemporal clusters. After calculating the coordinates of the cluster centers, ordinary kriging is used for spatial interpolation to finally obtain a continuous distribution of candidate locations of discharge sources.

[0080] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0081] (1) Divide the monitoring area into grids to obtain an antenna layout plan, and install a directional antenna array according to the antenna layout plan to obtain an antenna array network;

[0082] (2) Injecting a calibration signal into the antenna array network to obtain phase response data, and performing compensation calculation on the phase response data to obtain calibration parameters;

[0083] (3) Collect electromagnetic wave signals through the antenna array network to obtain multi-channel raw data, and perform phase comparison on the multi-channel raw data to obtain a phase difference matrix;

[0084] (4) Estimating the direction of arrival based on the phase difference matrix and calibration parameters to obtain the angle of arrival information, and then smoothing the angle of arrival information to obtain stable angle data;

[0085] (5) Perform multi-station intersection calculation on the stable angle data to obtain a candidate space point set, and perform least squares fitting on the candidate space point set to obtain the optimal fitting point;

[0086] (6) Perform coordinate transformation on the optimal fitting point to obtain the preliminary spatial coordinates.

[0087] Specifically, the monitoring area is gridded to obtain an antenna layout plan. The grid division adopts an equal spacing method to divide the three-dimensional space into several cubic units, and the side length of each unit is determined according to the positioning accuracy requirements. According to the divided grid, a suitable location is selected to install a directional antenna array to form an antenna array network. The layout of the antenna array needs to consider the coverage range and signal reception quality. Usually, a uniform distribution or optimization algorithm is used to determine the antenna position. A calibration signal is injected into the antenna array network to obtain phase response data. The calibration signal is usually transmitted using a standard signal source with a known position, and the signal frequency covers the typical frequency band of partial discharge. The phase response data records the phase information of each antenna receiving the calibration signal. Compensation calculation is performed on the phase response data to obtain calibration parameters. The purpose of the compensation calculation is to eliminate the phase differences between the channels in the antenna array. The calibration parameters are usually expressed as phase correction values. The calculation formula for the calibration parameters is:

[0088]

[0089] in, is the calibration parameter of the ith antenna, is the phase of the reference antenna, It is i The phase of each antenna is measured. Electromagnetic wave signals are collected through the antenna array network to obtain multi-channel raw data. During the collection process, each antenna synchronously receives the electromagnetic wave signals generated by partial discharge. Phase comparison is performed on the multi-channel raw data to obtain a phase difference matrix. Phase difference matrix The elements of are calculated as follows:

[0090]

[0091] in, is the phase difference between antennas i and j, and are the signal phases received by antenna i and j respectively, and is the corresponding calibration parameter. The direction of arrival is estimated based on the phase difference matrix and the calibration parameters to obtain the arrival angle information. The direction of arrival estimation uses the MUSIC (Multiple Signal Classification) algorithm, the core of which is to use the orthogonality of the signal subspace and the noise subspace. The spatial spectrum function of the MUSIC algorithm is:

[0092]

[0093] in, is the array manifold vector, is the eigenvector matrix of the noise subspace, and Represent the pitch angle and azimuth respectively. Perform data smoothing on the arrival angle information to obtain stable angle data. Data smoothing usually uses methods such as moving average or Kalman filtering to reduce the impact of instantaneous fluctuations. Perform multi-station intersection calculation on the stable angle data to obtain a candidate space point set. Multi-station intersection uses the principle of triangulation to determine the spatial intersection using the arrival angle information of different antenna stations. Perform least squares fitting on the candidate space point set to obtain the optimal fitting point. The objective function of least squares fitting is:

[0094]

[0095] in, is the observed distance, is the estimated distance, is the number of antennas. Finally, coordinate transformation is performed on the optimal fitting point to obtain preliminary spatial coordinates. Coordinate transformation transforms the local coordinate system into the global coordinate system for integration with other systems.

[0096] For example, in the local discharge location of a 500kV GIS device, the monitoring area size is 10m×8m×5m, and it is divided into a grid of 1m×1m×1m. According to the division results, 8 directional antennas are arranged around the equipment to form an antenna array network. The 500MHz standard signal source is used for calibration to obtain the phase response data of the 8 antennas. The calibration parameters are obtained through compensation calculation, and the maximum phase correction value is 15°. In actual monitoring, the antenna array simultaneously collects 1ms signals with a sampling rate of 1GS / s. The collected multi-channel data is phase contrasted to obtain 28 independent phase difference values ​​(8 out of 2 combinations). The MUSIC algorithm is used to estimate the direction of arrival and obtain the pitch angle and azimuth angle corresponding to each antenna. The angle data is smoothed by a 5-point moving average, which reduces the angle fluctuation by about 30%. The multi-station intersection calculation obtains 15 candidate spatial points, and the root mean square error of the least squares fitting is 0.12m. Finally, the optimal fitting point was converted to the global coordinate system to obtain the preliminary spatial coordinates (3.25m, 2.18m, 1.73m), which provided basic data for subsequent precise positioning and discharge source identification.

[0097] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0098] (1) Perform Monte Carlo simulation on the preliminary spatial coordinates to obtain coordinate distribution samples, and perform statistical analysis on the coordinate distribution samples to obtain error ellipse parameters;

[0099] (2) Calculate the positioning accuracy index based on the error ellipse parameters to obtain the accuracy evaluation matrix, and perform threshold segmentation on the accuracy evaluation matrix to obtain the credible area boundary;

[0100] (3) Analyze the spatial relationship between the credible area boundary and the candidate location of the discharge source to obtain the overlapping area, and perform density estimation on the overlapping area to obtain the probability density function;

[0101] (4) Perform maximum likelihood estimation based on the probability density function to obtain the most likely discharge source location, and perform a local search on the most likely discharge source location to obtain a refined location estimate;

[0102] (5) Multi-source information fusion is performed on the refined position estimation to obtain a comprehensive score, and the candidates are sorted and screened according to the comprehensive score to obtain a sorted candidate position set;

[0103] (6) Perform weighted averaging on the sorted candidate position set to obtain the optimized discharge source position.

[0104] Specifically, the preliminary spatial coordinates are randomly perturbed multiple times to generate coordinate distribution samples. This process is achieved by adding random noise that obeys the normal distribution. The number of perturbations is usually set to more than 1000 times to ensure the statistical significance of the samples. The generated coordinate distribution samples are statistically analyzed, and the mean and covariance matrix of the samples are calculated to obtain the error ellipse parameters. The error ellipse parameters include the major axis, minor axis and direction angle of the ellipse, which intuitively reflect the uncertainty of the positioning results. According to the error ellipse parameters, the positioning accuracy index is calculated. The main indicators include the horizontal positioning error circular probability (CEP) and the vertical positioning error probability (VEP). These indicators constitute the accuracy assessment matrix, and the matrix is ​​segmented by setting a threshold to obtain the credible area boundary. The selection of the threshold is usually based on the accuracy requirements of the actual application. For example, the error range at a 95% confidence level can be selected as the credible area boundary.

[0105] Analyze the spatial relationship between the credible area boundary and the candidate location of the discharge source to find the overlapping area between the two. This is achieved through computational geometry methods, such as the polygon intersection algorithm. Density estimation is performed on the overlapping area to obtain a probability density function. Density estimation methods can choose non-parametric methods such as kernel density estimation or mixed Gaussian models, which can flexibly describe complex spatial distributions. Based on the obtained probability density function, maximum likelihood estimation is performed to determine the most likely location of the discharge source. The goal of maximum likelihood estimation is to find the parameter value that maximizes the probability of the observed data. On this basis, a local search is performed on the most likely location of the discharge source to obtain a more precise location estimate. Local search can use optimization algorithms such as gradient descent or simulated annealing to find a better solution in a small range near the most likely location.

[0106] Multi-source information fusion is performed on the refined position estimation, and various information such as electromagnetic field strength, arrival time difference, phase difference, etc. are comprehensively considered to obtain a comprehensive score. Multi-source information fusion can use methods such as evidence theory or fuzzy reasoning to organically combine information from different sources. Sorting and screening are performed according to the comprehensive score to obtain a set of sorted candidate positions. The sorting process takes into account the weights of various indicators to ensure the reliability of the final result. Finally, the sorted candidate position set is weighted averaged to obtain the optimized discharge source position. When weighted averaging, the weight can be determined based on the comprehensive score of each candidate position, and the position with a higher score obtains a greater weight. This method can make full use of the information of multiple high-quality candidate positions and improve the accuracy and stability of the final positioning result.

[0107] For example, in the process of localizing partial discharge of a 500kV GIS device, the initial spatial coordinates are (10.5m, 5.2m, 2.8m). 10,000 Monte Carlo simulations were performed, and Gaussian noise with a mean of 0 and a standard deviation of 0.1m was added in three directions each time. Statistical analysis obtained the error ellipse parameters: major axis 0.25m, minor axis 0.18m, and direction angle 30°. The calculated CEP is 0.23m and the VEP is 0.15m. Setting a 95% confidence level, the boundary of the credible area is an ellipsoid, and its horizontal cross section is an ellipse with a major axis of 0.49m and a minor axis of 0.35m.

[0108] The spatial relationship analysis with the candidate locations of the discharge source was performed, and it was found that three candidate locations fell within the credible area. The kernel density estimation was performed on these three locations to obtain a three-dimensional probability density function. The maximum likelihood estimation obtained the most likely discharge source location as (10.48m, 5.23m, 2.79m). A local search was performed within 0.5m of this location, and the simulated annealing algorithm was used to obtain a refined location estimate (10.47m, 5.22m, 2.81m). Multi-source information fusion took into account the electromagnetic field strength (weight 0.4), arrival time difference (weight 0.4) and phase difference (weight 0.2), and the comprehensive scores of the three candidate locations were 0.92, 0.88 and 0.85 respectively. The weighted average of these three locations was performed, and the weight was proportional to the score, and finally the optimized discharge source location (10.475m, 5.225m, 2.805m) was obtained. This result improves the accuracy compared to the initial estimate, with the horizontal error reduced from the initial 0.23m to 0.13m and the vertical error reduced from 0.15m to 0.09m, significantly improving the accuracy of positioning.

[0109] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0110] (1) The optimized discharge source location is segmented into time windows to obtain time series data segments, and the time series data segments are Fourier transformed to obtain frequency domain feature sequences;

[0111] (2) Extract the peak value of the frequency domain feature sequence to obtain a feature frequency set, and perform pattern matching on the feature frequency set to obtain a discharge frequency template;

[0112] (3) Perform time domain reconstruction based on the discharge frequency template to obtain a typical discharge waveform, and then perform parameterized description of the typical discharge waveform to obtain a discharge characteristic parameter set;

[0113] (4) Perform time series analysis on the discharge characteristic parameter set to obtain discharge pattern data, and perform cluster analysis on the discharge pattern data to obtain the discharge pattern category;

[0114] (5) Construct a decision tree model based on the discharge pattern category to obtain a preliminary determination result of the discharge type, and perform fuzzy reasoning on the preliminary determination result of the discharge type to obtain the probability distribution of the discharge type;

[0115] (6) Comprehensively evaluate the discharge type probability distribution and discharge characteristic parameter set to obtain the discharge type and severity assessment results.

[0116] It should be noted that the continuous time series data is segmented at fixed time intervals to obtain multiple time series data segments. This segmentation process usually uses a sliding window technique, and the window size is determined according to the characteristics of the discharge signal, for example, it can be set to 1ms. Fourier transform is performed on each time series data segment, and the time domain signal is converted into a frequency domain representation to obtain a frequency domain feature sequence. This step helps to analyze the frequency characteristics of the discharge signal and reveal the characteristic frequency distribution of different types of discharges. Peak extraction is performed on the frequency domain feature sequence to obtain a characteristic frequency set. Peak extraction uses a local maximum search algorithm and sets a threshold to filter noise. The selection of the threshold is usually based on the average energy level of the signal, for example, it can be set to 3 times the average energy. Pattern matching is performed on the characteristic frequency set to obtain a discharge frequency template. The pattern matching process uses a dynamic time warping algorithm to calculate the similarity between the characteristic frequency set and the preset template. The characteristic frequency patterns of different types of discharges can be identified.

[0117] The time domain is reconstructed according to the discharge frequency template to obtain a typical discharge waveform. The reconstruction process uses inverse Fourier transform to convert the frequency domain information back to the time domain. The typical discharge waveform is parameterized and features such as peak amplitude, rise time, and duration are extracted to obtain a set of discharge feature parameters. These parameters intuitively reflect the time domain characteristics of the discharge signal.

[0118] Perform time series analysis on the discharge feature parameter set and use the autoregressive moving average model to obtain the discharge pattern data. It can capture the time dynamic characteristics of discharge behavior, such as periodicity and trend. Perform cluster analysis on the discharge pattern data and use K-means or hierarchical clustering algorithm to obtain the discharge pattern category. Cluster analysis helps to identify different types of discharge patterns. Construct a decision tree model based on the discharge pattern category to obtain the preliminary judgment result of the discharge type. The construction of the decision tree is based on the information gain criterion, and the most discriminative feature is selected as the split node. Fuzzy reasoning is performed on the preliminary judgment result of the discharge type to obtain the probability distribution of the discharge type. Fuzzy reasoning uses the Mamdani model to convert expert experience into a fuzzy rule set.

[0119] Finally, the discharge type probability distribution and the discharge characteristic parameter set are comprehensively evaluated to obtain the discharge type and severity evaluation results. The comprehensive evaluation adopts the weighted summation method, and the weights are determined according to the importance of different parameters.

[0120] For example, in the partial discharge detection of a 500kV GIS device, the discharge source position obtained by optimized positioning is analyzed. First, the continuous signal of 1 second is divided into time windows, the window size is set to 1ms, the step size is 0.5ms, and 2000 time series data fragments are obtained. Each fragment is subjected to fast Fourier transform to obtain the frequency domain feature sequence. In the frequency domain features, it is found that the main energy is concentrated in the range of 300-500MHz. The peak extraction threshold is set to 3 times the average energy, and 5 main feature frequencies are extracted: 350MHz, 400MHz, 450MHz, 480MHz and 500MHz. By dynamically time-warping matching with the preset discharge type template, the best matching discharge frequency template is obtained, and the matching degree is 0.92. The time domain waveform is reconstructed according to this template to obtain a typical discharge waveform. The features obtained by parameterized description include: peak amplitude of 80mV, rise time of 2ns, and duration of 10ns. Time series analysis shows that the discharge behavior is periodic, with a period of about 10ms, which is related to the grid frequency. Cluster analysis divides the discharge mode into three categories, and the decision tree model preliminarily determines it as suspended potential discharge. The probability distribution of discharge types obtained by fuzzy reasoning is: suspended potential discharge 80%, particle discharge 15%, and other types 5%. Comprehensive evaluation takes into account factors such as discharge amplitude, frequency, and duration, and ultimately determines it to be a medium-severity suspended potential discharge. It is recommended to strengthen monitoring and handle it during the next maintenance.

[0121] The embodiment of the present invention also provides a partial discharge location system based on radio waves, such as Figure 2 As shown, the local discharge locating system based on radio waves specifically includes:

[0122] The acquisition module 201 is used to acquire and process the multi-band broadband signals to obtain raw electromagnetic signal data, and to pre-process and reduce noise on the raw electromagnetic signal data to obtain cleaned signal data;

[0123] The analysis module 202 is used to perform time-frequency analysis on the cleaning signal data to obtain time-frequency feature data, and perform feature extraction on the time-frequency feature data to obtain a partial discharge feature vector;

[0124] A clustering module 203 is used to perform cluster analysis on the local discharge feature vectors to obtain a discharge event clustering result, and perform spatiotemporal correlation analysis on the discharge event clustering result to obtain a discharge source candidate location;

[0125] The calculation module 204 is used to arrange multiple directional antenna arrays in the monitoring area, collect electromagnetic wave direction data in real time, obtain arrival angle information, and perform triangulation positioning calculation on the arrival angle information to obtain preliminary spatial coordinates;

[0126] A matching module 205 is used to perform error analysis on the preliminary spatial coordinates to obtain positioning accuracy evaluation data, and to perform matching analysis on the positioning accuracy evaluation data and the candidate discharge source position to obtain an optimized discharge source position;

[0127] The identification module 206 is used to perform a time series analysis on the optimized discharge source position to obtain discharge pattern data, and to perform intelligent identification on the discharge pattern data to obtain a discharge type and severity evaluation result.

[0128] Through the collaborative work of the above modules, the quality of the original electromagnetic signal is effectively improved by collecting and processing the multi-band broadband signal and pre-processing the noise reduction. The time-frequency analysis and feature extraction of the cleaning signal data can not only fully capture the time-frequency characteristics of the partial discharge signal, but also effectively compress the data volume and improve the processing efficiency. By clustering and spatiotemporal correlation analysis of the partial discharge feature vector, different types of discharge events can be effectively identified and classified, and the possible location of the discharge source can be preliminarily determined, greatly narrowing the search range. Multiple directional antenna arrays are arranged in the monitoring area to collect electromagnetic wave direction data in real time. Combined with triangulation positioning calculation, the preliminary spatial coordinates of the discharge source can be quickly and accurately determined. Error analysis and matching analysis of the preliminary spatial coordinates can not only evaluate the reliability of the positioning results, but also further optimize the positioning results by comparing with the candidate positions of the discharge source, significantly improving the positioning accuracy. Finally, the timing analysis and intelligent identification of the optimized discharge source position can not only extract the discharge mode data, but also accurately judge the discharge type and severity. Through multi-step data processing and analysis, the whole process from signal acquisition to discharge source positioning to discharge type identification is realized, which greatly improves the efficiency and accuracy of partial discharge detection.

[0129] 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 embodiments, a person skilled in the art should understand that the specific implementation modes of the present invention can still be modified or replaced by equivalents, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for locating partial discharge based on radio waves, characterized in that: include: The multi-band broadband signal is collected and processed to obtain original electromagnetic signal data, and the original electromagnetic signal data is preprocessed and denoised to obtain cleaned signal data, including: adaptively sampling the multi-band broadband signal to obtain original sampling data, and performing spectrum analysis on the original sampling data to obtain frequency distribution characteristics; band-pass filtering the frequency distribution characteristics to obtain target frequency band signals, and reconstructing the target frequency band signals in time domain to obtain time domain reconstructed signals; performing wavelet transform on the time domain reconstructed signals to obtain wavelet coefficients, and performing threshold denoising on the wavelet coefficients to obtain denoised wavelet coefficients; reconstructing the denoised wavelet coefficients by inverse wavelet transform to obtain a preliminary denoised signal, and performing empirical mode decomposition on the preliminary denoised signal to obtain an eigenmode function set; performing correlation analysis on the eigenmode function set to obtain main modal components, and reconstructing and superimposing the main modal components to obtain a reconstructed signal; performing mean filtering on the reconstructed signal to obtain a smoothed signal, and normalizing the smoothed signal to obtain cleaned signal data; Performing time-frequency analysis on the cleaning signal data to obtain time-frequency feature data, and performing feature extraction on the time-frequency feature data to obtain a local discharge feature vector; Performing cluster analysis on the local discharge feature vectors to obtain a discharge event clustering result, and performing spatiotemporal correlation analysis on the discharge event clustering result to obtain a discharge source candidate location; Arrange multiple directional antenna arrays in the monitoring area to collect electromagnetic wave direction data in real time to obtain arrival angle information, and perform triangulation positioning calculation on the arrival angle information to obtain preliminary spatial coordinates; Performing error analysis on the preliminary spatial coordinates to obtain positioning accuracy evaluation data, and performing matching analysis on the positioning accuracy evaluation data and the candidate discharge source positions to obtain an optimized discharge source position; The optimized discharge source position is subjected to a time series analysis to obtain discharge pattern data, and the discharge pattern data is subjected to intelligent identification to obtain a discharge type and severity evaluation result.

2. The method for locating partial discharge based on radio waves according to claim 1, characterized in that: The step of performing time-frequency analysis on the cleaning signal data to obtain time-frequency feature data, and performing feature extraction on the time-frequency feature data to obtain a partial discharge feature vector comprises: Performing short-time Fourier transform on the cleaning signal data to obtain a time-frequency spectrum, and performing energy density calculation on the time-frequency spectrum to obtain an energy distribution matrix; Performing peak detection on the energy distribution matrix to obtain a characteristic peak point set, and performing cluster analysis on the characteristic peak point set to obtain a main frequency band interval; Performing time series slicing on the main frequency band interval to obtain a time-frequency slice set, and performing wavelet packet decomposition on the time-frequency slice set to obtain wavelet packet coefficients; Calculating the entropy value of the wavelet packet coefficients to obtain information entropy features, and performing statistical analysis on the information entropy features to obtain entropy value distribution features; Performing dimensionality reduction processing on the entropy value distribution feature through a principal component analysis algorithm to obtain a low-dimensional feature representation, and performing nonlinear mapping on the low-dimensional feature representation to obtain a high-order feature; Feature selection is performed on the high-order features to obtain a key feature subset, and feature fusion is performed on the key feature subset to obtain a partial discharge feature vector.

3. The method for locating partial discharge based on radio waves according to claim 1, characterized in that: The step of performing cluster analysis on the local discharge feature vectors to obtain a discharge event clustering result, and performing spatiotemporal correlation analysis on the discharge event clustering result to obtain a discharge source candidate location comprises: Performing density peak clustering on the partial discharge feature vector to obtain an initial clustering result, and performing silhouette coefficient evaluation on the initial clustering result to obtain a clustering quality index; Adaptively adjusting clustering parameters according to the clustering quality index to obtain optimized clustering parameters, and re-executing density peak clustering using the optimized clustering parameters to obtain discharge event clustering results; Performing timestamp analysis on the discharge event clustering results to obtain time series features, and performing periodicity detection on the time series features to obtain a discharge periodicity pattern; Performing spatial distribution analysis on the clustering results of the discharge events to obtain a spatial density map, and performing hot spot identification on the spatial density map to obtain coordinates of high-incidence areas; Constructing a spatiotemporal association network according to the discharge cycle pattern and the coordinates of the high-incidence area to obtain an association weight matrix, and performing spectral clustering on the association weight matrix to obtain a spatiotemporal cluster cluster; The centroid of the spatiotemporal clusters is calculated to obtain a cluster center coordinate set, and the cluster center coordinate set is spatially interpolated to obtain a candidate location of a discharge source.

4. The method for locating partial discharge based on radio waves according to claim 1, characterized in that: The step of arranging a plurality of directional antenna arrays in the monitoring area, collecting electromagnetic wave direction data in real time, obtaining arrival angle information, and performing triangulation positioning calculation on the arrival angle information to obtain preliminary spatial coordinates includes: Dividing the monitoring area into grids to obtain an antenna arrangement plan, and installing a directional antenna array according to the antenna arrangement plan to obtain an antenna array network; Injecting a calibration signal into the antenna array network to obtain phase response data, and performing compensation calculation on the phase response data to obtain calibration parameters; Collecting electromagnetic wave signals through the antenna array network to obtain multi-channel raw data, and performing phase comparison on the multi-channel raw data to obtain a phase difference matrix; Estimating the direction of arrival according to the phase difference matrix and the calibration parameters to obtain angle of arrival information, and performing data smoothing on the angle of arrival information to obtain stable angle data; Performing multi-station intersection calculation on the stable angle data to obtain a candidate space point set, and performing least square fitting on the candidate space point set to obtain an optimal fitting point; The optimal fitting point is subjected to coordinate transformation to obtain preliminary space coordinates.

5. The method for locating partial discharge based on radio waves according to claim 1, characterized in that: The step of performing error analysis on the preliminary spatial coordinates to obtain positioning accuracy evaluation data, and performing matching analysis on the positioning accuracy evaluation data and the candidate discharge source positions to obtain an optimized discharge source position comprises: Performing Monte Carlo simulation on the preliminary spatial coordinates to obtain coordinate distribution samples, and performing statistical analysis on the coordinate distribution samples to obtain error ellipse parameters; Calculating the positioning accuracy index according to the error ellipse parameters to obtain an accuracy evaluation matrix, and performing threshold segmentation on the accuracy evaluation matrix to obtain a credible area boundary; Performing spatial relationship analysis on the boundary of the credible region and the candidate position of the discharge source to obtain an overlapping region, and performing density estimation on the overlapping region to obtain a probability density function; Performing maximum likelihood estimation according to the probability density function to obtain the most likely discharge source location, and performing local search on the most likely discharge source location to obtain a refined location estimate; Perform multi-source information fusion on the refined position estimation to obtain a comprehensive score, and perform sorting and screening according to the comprehensive score to obtain a sorted candidate position set; The sorted candidate position set is weighted averaged to obtain an optimized discharge source position.

6. The method for locating partial discharge based on radio waves according to claim 5, characterized in that: The step of performing a time series analysis on the optimized discharge source position to obtain discharge mode data, and performing intelligent identification on the discharge mode data to obtain a discharge type and severity assessment result comprises: Performing time window segmentation on the optimized discharge source position to obtain time series data segments, and performing Fourier transform on the time series data segments to obtain frequency domain feature sequences; Performing peak extraction on the frequency domain feature sequence to obtain a feature frequency set, and performing pattern matching on the feature frequency set to obtain a discharge frequency template; Performing time domain reconstruction according to the discharge frequency template to obtain a typical discharge waveform, and performing parameterized description on the typical discharge waveform to obtain a discharge characteristic parameter set; Performing time series analysis on the discharge characteristic parameter set to obtain discharge pattern data, and performing cluster analysis on the discharge pattern data to obtain discharge pattern categories; Constructing a decision tree model according to the discharge mode category to obtain a preliminary determination result of the discharge type, and performing fuzzy reasoning on the preliminary determination result of the discharge type to obtain a probability distribution of the discharge type; The discharge type probability distribution and the discharge characteristic parameter set are comprehensively evaluated to obtain a discharge type and severity evaluation result.

7. A radio wave based partial discharge location system, used to execute the radio wave based partial discharge location method according to any one of claims 1 to 6, characterized in that: include: The acquisition module is used to acquire and process the multi-band broadband signal to obtain the original electromagnetic signal data, and pre-process and reduce noise on the original electromagnetic signal data to obtain the cleaned signal data, including: adaptively sampling the multi-band broadband signal to obtain the original sampling data, and performing spectrum analysis on the original sampling data to obtain the frequency distribution characteristics; band-pass filtering the frequency distribution characteristics to obtain the target frequency band signal, and reconstructing the target frequency band signal in the time domain to obtain the time domain reconstructed signal; performing wavelet transform on the time domain reconstructed signal to obtain wavelet coefficients, and performing threshold denoising on the wavelet coefficients to obtain denoised wavelet coefficients; reconstructing the denoised wavelet coefficients by inverse wavelet transform to obtain a preliminary denoised signal, and performing empirical mode decomposition on the preliminary denoised signal to obtain an eigenmode function set; performing correlation analysis on the eigenmode function set to obtain the main modal components, and reconstructing and superimposing the main modal components to obtain the reconstructed signal; performing mean filtering on the reconstructed signal to obtain a smoothed signal, and normalizing the smoothed signal to obtain the cleaned signal data; An analysis module, used for performing time-frequency analysis on the cleaning signal data to obtain time-frequency feature data, and performing feature extraction on the time-frequency feature data to obtain a local discharge feature vector; A clustering module, used for performing cluster analysis on the local discharge feature vectors to obtain a discharge event clustering result, and performing spatiotemporal correlation analysis on the discharge event clustering result to obtain a discharge source candidate location; A calculation module is used to arrange multiple directional antenna arrays in the monitoring area, collect electromagnetic wave direction data in real time, obtain arrival angle information, and perform triangulation positioning calculation on the arrival angle information to obtain preliminary spatial coordinates; A matching module, used for performing error analysis on the preliminary spatial coordinates to obtain positioning accuracy evaluation data, and performing matching analysis on the positioning accuracy evaluation data and the candidate position of the discharge source to obtain an optimized discharge source position; The identification module is used to perform a time series analysis on the optimized discharge source position to obtain discharge mode data, and to perform intelligent identification on the discharge mode data to obtain a discharge type and severity evaluation result.

Images