Partial discharge online monitoring method based on ultrahigh frequency original signal

Abstract

The invention discloses a partial discharge online monitoring method based on an ultrahigh frequency original signal, and relates to the field of partial discharge monitoring, and the method comprises the steps: autonomously switching an ultrahigh frequency signal capturing frequency band according to the real-time electromagnetic spectrum distribution of a monitoring environment, and obtaining an original signal sequence covering a potential partial discharge feature; the method comprises the following steps: preprocessing an original signal sequence through multi-scale pulse contour extraction, and suppressing environmental noise interference to retain pulse characteristics of partial discharge signals; through real-time electromagnetic spectrum analysis, ultrahigh frequency capture frequency bands are automatically switched, environmental noise changes are dynamically adapted, original signal capture accuracy is improved, through multi-scale pulse contour extraction preprocessing, noise interference is effectively suppressed, key pulse features are reserved, and the accuracy of signal capture is improved. A multi-dimensional feature data set constructed based on the pulse peak gradient, the duration spectrum and the waveform similarity provides a comprehensive basis for partial discharge recognition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of partial discharge monitoring, in particular to a partial discharge online monitoring method based on ultra-high frequency original signals. BACKGROUND

[0002] The partial discharge monitoring technology based on ultra-high frequency original signals is to capture the ultra-high frequency electromagnetic waves generated by the partial discharge of the equipment, collect and analyze the original signals in real time, accurately identify insulation defects, realize early fault warning of high-voltage equipment, and ensure the safe operation of the power system.

[0003] The invention patent application with the application number 202411292974.1 discloses a GIS ultra-high frequency partial discharge online detection equipment dynamic range detection method, which comprises: performing signal processing on the real partial discharge signal by a statistical method to generate a representative partial discharge signal as an injection signal; testing whether the output signals of an arbitrary waveform generator for the injection signal are distorted; building a GIS ultra-high frequency partial discharge online monitoring system to measure the dynamic range of the GIS ultra-high frequency partial discharge online monitoring equipment using the undistorted injection signal; determining the background noise of the built GIS ultra-high frequency partial discharge online monitoring system to determine the dynamic range of the GIS ultra-high frequency partial discharge online monitoring equipment at a specified frequency point until the dynamic range of the GIS ultra-high frequency partial discharge online monitoring equipment at different frequencies is obtained. The application aims to solve the problem of "lack of dynamic range detection and testing method for ultra-high frequency partial discharge online monitoring equipment in the prior art".

[0004] However, for the partial discharge defects of the equipment, the prior art mostly focuses on the detection of partial discharge problems, and the positioning accuracy is poor after the partial discharge problem is monitored.

[0005] Therefore, a partial discharge online monitoring method based on ultra-high frequency original signals is proposed. SUMMARY

[0006] In view of the above shortcomings of the prior art, the present application provides a partial discharge online monitoring method based on ultra-high frequency original signals, which can effectively solve the problems of the prior art.

[0007] To achieve the above purpose, the present application is realized by the following technical scheme.

[0008] The present application discloses a partial discharge online monitoring method based on ultra-high frequency original signals, which comprises:

[0009] Based on the real-time electromagnetic spectrum distribution of the monitoring environment, the system autonomously switches the UHF signal acquisition band to obtain the original signal sequence covering potential partial discharge (PD) characteristics. The original signal sequence is preprocessed using multi-scale pulse profile extraction to suppress environmental noise interference and preserve the pulse characteristics of the PD signal. Three spatiotemporal distribution parameters—pulse peak gradient, duration spectrum, and waveform similarity—are extracted from the preprocessed signal to construct a multidimensional PD characteristic dataset. A dynamic benchmark comparison model is created to quantify the deviation between the multidimensional PD characteristic dataset and the dynamic benchmark, identifying the activity level and type of the PD signal. Based on the time difference and waveform feature matching degree of the identified PD signals arriving at each monitoring node, the spatial coordinates of the PD source are obtained through distributed spatiotemporal correlation calculation. Based on the historical PD characteristic change rate and the current multidimensional PD characteristic dataset, a development trend prediction model is constructed, and based on the model, hierarchical early warning information including location, intensity, and risk level is synchronously output.

[0010] Furthermore, the specific process of autonomously switching the UHF signal acquisition frequency band based on the real-time electromagnetic spectrum distribution of the monitoring environment includes:

[0011] Real-time acquisition of broadband electromagnetic spectrum data of the monitoring environment, and calculation of the noise energy proportion of each preset ultra-high frequency sub-band. and signal energy ratio And determine the optimal acquisition frequency band based on the following formula;

[0012] ;

[0013] In the formula: This is the optimal ultra-high frequency signal capture band; This is a pre-defined set of UHF candidate frequency bands; This represents the percentage of signal energy within frequency band f at time t. This represents the percentage of noise energy in frequency band f at time t. This is the frequency band stability factor, with a value range of [0,1].

[0014] in, The value follows this rule: when the ratio of the standard deviation to the mean of the frequency band energy is less than or equal to 0.15. =1; when the ratio of the standard deviation to the mean of the frequency band energy is greater than 0.15, The frequency band decreases linearly as the ratio increases, and frequency band switching is performed when the signal energy ratio of the optimal frequency band for two consecutive sampling periods is not less than 0.6.

[0015] Furthermore, the multi-scale pulse contour extraction preprocessing operation of the original signal sequence follows the following:

[0016] Variational mode decomposition is used to decompose the original signal sequence x(t) into K modal components. ;

[0017] Calculate the impulse significance index for each mode. ;

[0018] ;

[0019] In the formula: The absolute maximum value of the k-th modal component signal; The mean of the absolute values ​​of the k-th modal component signal; Let V be the variance of the absolute value of the k-th modal component signal;

[0020] Among them, retain ≥ The modal components are analyzed, and adaptive threshold denoising is applied to the retained components to reconstruct the preprocessed signal. The preset threshold is set to 3.5; the threshold value is used in the adaptive threshold denoising process. obey:

[0021] , This represents the median of the absolute values ​​of the k-th modal component signal.

[0022] Furthermore, the operation of extracting the pulse peak gradient includes:

[0023] Peak detection is performed on the preprocessed signal to obtain the peak amplitude sequence. and corresponding time , This indicates the total number of detected pulse peaks;

[0024] Define the gradient of adjacent peaks :

[0025] ;

[0026] In the formula: , Let be the amplitude of the peak values ​​of the p-th and (p-1)-th pulses; , These are the times when the peak values ​​of the p-th and (p-1)-th pulses occur; To find the minimum value, take ;

[0027] Among them, ≥ The gradient values ​​are included in the feature dataset. This indicates a preset threshold that is dynamically adjusted based on the historical gradient mean.

[0028] Furthermore, the extraction step of the duration spectrum includes:

[0029] With sliding time window , The window width is represented by 1.5 times the average pulse period. The preprocessed signal is truncated using a sliding method, and the pulse duration within the window is calculated. ;

[0030] Constructing the duration spectrum :

[0031] ;

[0032] extract The spectral peak positions and corresponding amplitudes are used as characteristic parameters;

[0033] In the formula: This is the time-domain sequence of the preprocessed partial discharge signal; For the m-th sliding time window, This represents the total number of sliding time windows; The Dirac delta function only when The value is infinity when it is infinity, and 0 in other cases, and satisfies the following conditions: ; This is a reference value for the typical duration of a partial discharge pulse; It is a time variable; The horizontal axis variable of the duration spectrum represents a general variable indicating the duration of the partial discharge pulse from start to end, covering all possible pulse duration values.

[0034] Furthermore, the process of extracting waveform similarity includes:

[0035] The pulse waveforms with the highest amplitude in the preprocessed signal were selected as candidate templates, and the baseline template was obtained by screening using cross-correlation coefficients. ;

[0036] For any waveform to be compared After time alignment, similarity is calculated. ;

[0037] ;

[0038] In the formula: The duration of the waveform; For weighted functions, t is a time variable, representing consecutive moments on the time axis;

[0039] Will Waveforms with a value ≥0.75 are marked as similar waveforms, and the proportion of similarity is statistically analyzed and included in the feature dataset.

[0040] Furthermore, the creation of the dynamic benchmark comparison model follows:

[0041] A baseline feature matrix is ​​constructed based on the most recent N normal state samples. Calculation benchmark center and feature weight matrix ;

[0042] The deviation D between the current eigenvector X and the benchmark is ;

[0043] The new benchmark is updated using a sliding window. ;

[0044] When D≥ At that time, according to D and The comparison is used to classify the localized activity level;

[0045] In the formula: This is the multidimensional feature vector of the Nth normal state sample; This is a transpose operation for a matrix or vector. Let be the standard deviation of the k-th feature. This is the reference center vector before the update, i.e., the reference center of the previous period; This is the preset deviation threshold.

[0046] Furthermore, the process for obtaining the spatial coordinates of the partial discharge source through distributed spatiotemporal correlation calculation includes:

[0047] Let the coordinates of the M monitoring nodes be... The signal arrival time is Define the source coordinates (x, y, z) of the partial discharge source to satisfy:

[0048] ;

[0049] In the formula: For signal propagation speed; This refers to the moment when partial discharge occurs; This is a correction term for propagation distance;

[0050] The spatial coordinates of the partial discharge source are obtained by solving the system of equations using the least squares method.

[0051] in, , This represents the environmental attenuation coefficient, ranging from 0.01 to 0.05. It is applied when the density of metal obstacles ρ in the monitored environment is ≥0.3 objects / m³ or the relative humidity H is ≥75%. Take a value of 0.03~0.05; when the density of metal obstacles ρ < 0.2 objects / m³ and the relative humidity H < 60%, Take 0.01~0.02; under other environmental conditions Take a value of 0.02~0.03; This represents the straight-line distance from the node to the source.

[0052] Furthermore, the development trend prediction model is as follows:

[0053] Based on a historical partial discharge feature dataset spanning T periods, the feature growth rate is calculated. ;

[0054] Constructing the risk trend index R:

[0055] ;

[0056] In the formula: Let be the multidimensional partial discharge eigenvector of the T-th period; This is the timestamp for the Tth period; These are the weighting coefficients; This represents the deviation between the current feature vector X and the baseline. It is the average of the absolute values ​​of the differences between all adjacent periodic eigenvectors;

[0057] Specifically, when R < R1, the output is low risk; when R1 ≤ R < R2, the output is medium risk; and when R ≥ R2, the output is high risk. R1 and R2 are preset thresholds. The contribution ratios of the characteristic growth rate, deviation value, and historical volatility mean in the risk trend index are adjusted, initially set to 0.3, 0.5, and 0.2, respectively, and their sum is always 1.

[0058] Furthermore, the tiered early warning information also includes a partial discharge type feature matching score, the calculation process of which is as follows:

[0059] The extracted multidimensional partial discharge feature dataset is compared with a preset partial discharge type feature library in multiple dimensions, and the type matching score is calculated. :

[0060] ;

[0061] In the formula: For feature dimensions; The type weight of the current i-th feature; This represents the current i-th eigenvalue; Let i be the standard value of the i-th feature of the t-th type of partial discharge; , Let be the maximum and minimum values ​​of the normal fluctuations of the i-th characteristic of the t-th type of partial discharge; To find the minimum value, take ;

[0062] Among them, the early warning information simultaneously outputs the partial discharge type with the highest matching score and its corresponding... numerical value, when A value ≥0.65 indicates a valid type match.

[0063] The value of is determined based on the degree of dispersion of the feature among different local discharge types. The more dispersed the value is among different types, the larger the value will be, and vice versa.

[0064] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects:

[0065] This invention provides an online partial discharge (PD) monitoring method based on UHF raw signals. During execution, the method autonomously switches UHF acquisition bands through real-time electromagnetic spectrum analysis, dynamically adapting to changes in environmental noise and improving the accuracy of raw signal acquisition. Multi-scale pulse profile extraction preprocessing effectively suppresses noise interference while preserving key pulse features. A multi-dimensional feature dataset constructed based on pulse peak gradient, duration spectrum, and waveform similarity provides comprehensive evidence for PD identification. A dynamic benchmark comparison model accurately identifies the activity level and type of PD through deviation quantification. Distributed spatiotemporal correlation calculation combined with propagation correction terms achieves precise PD source localization. A development trend prediction model integrates historical feature changes with current data, outputting a graded early warning including location, intensity, risk level, and type matching scores. Weighting coefficients dynamically adjust the risk contribution ratio to ensure the scientific nature of the early warning. Overall, this method optimizes the entire process of PD monitoring, from signal acquisition and feature extraction to identification, localization, and risk warning, improving monitoring reliability and early warning capabilities. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0067] Figure 1 This is a flowchart illustrating an online partial discharge monitoring method based on ultra-high frequency raw signals. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0069] The present invention will be further described below with reference to embodiments.

[0070] Example:

[0071] This embodiment presents an online partial discharge monitoring method based on ultra-high frequency raw signals, such as... Figure 1 As shown, it includes:

[0072] Based on the real-time electromagnetic spectrum distribution of the monitoring environment, the system autonomously switches the UHF signal acquisition frequency band to obtain the original signal sequence covering potential partial discharge characteristics.

[0073] The specific process for autonomously switching the UHF signal acquisition frequency band based on the real-time electromagnetic spectrum distribution of the monitoring environment includes:

[0074] Real-time acquisition of broadband electromagnetic spectrum data of the monitoring environment, and calculation of the noise energy proportion of each preset ultra-high frequency sub-band. and signal energy ratio And determine the optimal acquisition frequency band based on the following formula;

[0075] ;

[0076] In the formula: This is the optimal ultra-high frequency signal capture band; This is a pre-defined set of UHF candidate frequency bands; This represents the percentage of signal energy within frequency band f at time t. This represents the percentage of noise energy in frequency band f at time t. This is the frequency band stability factor, with a value range of [0,1].

[0077] in, The value follows this rule: when the ratio of the standard deviation to the mean of the frequency band energy is less than or equal to 0.15. =1; when the ratio of the standard deviation to the mean of the frequency band energy is greater than 0.15, The frequency band decreases linearly as the ratio increases, and frequency band switching is performed when the signal energy ratio of the optimal frequency band for two consecutive sampling periods is not less than 0.6.

[0078] The above formula integrates the signal energy ratio, noise energy ratio, and frequency band stability factor to dynamically screen candidate frequency bands. It prioritizes the retention of frequency bands with high signal ratio and low noise interference, while avoiding frequency bands with drastic energy fluctuations through the stability factor. When the signal energy ratio of the optimal frequency band continuously meets the standard, it triggers switching to achieve adaptive tracking of changes in the electromagnetic spectrum of the monitoring environment.

[0079] The original signal sequence is preprocessed by multi-scale pulse profile extraction to suppress environmental noise interference and preserve the pulse characteristics of the partial discharge signal.

[0080] The multi-scale pulse contour extraction preprocessing operation of the original signal sequence follows the following rules:

[0081] Variational mode decomposition is used to decompose the original signal sequence x(t) into K modal components. ;

[0082] Calculate the impulse significance index for each mode. ;

[0083] ;

[0084] In the formula: The absolute maximum value of the k-th modal component signal; The mean of the absolute values ​​of the k-th modal component signal; Let V be the variance of the absolute value of the k-th modal component signal;

[0085] Among them, retain ≥ The modal components are analyzed, and adaptive threshold denoising is applied to the retained components to reconstruct the preprocessed signal. The preset threshold is set to 3.5; the threshold value is used in the adaptive threshold denoising process. obey:

[0086] The above formula constructs a quantitative index by using the absolute maximum value, mean, and variance of the modal component signals. By setting a threshold, modal components with significant pulse characteristics are selected. Combined with an adaptive threshold based on the median, the retained components are denoised. While suppressing noise interference, the pulse contour of the partial discharge signal is accurately preserved, laying a reliable foundation for subsequent feature extraction.

[0087] , This represents the median of the absolute values ​​of the k-th modal component signal;

[0088] Three spatiotemporal distribution parameters—pulse peak gradient, duration spectrum, and waveform similarity—are extracted from the preprocessed signal to construct a multidimensional partial discharge feature dataset.

[0089] The operations for extracting the pulse peak gradient include:

[0090] Peak detection is performed on the preprocessed signal to obtain the peak amplitude sequence. and corresponding time , This indicates the total number of detected pulse peaks;

[0091] Define the gradient of adjacent peaks :

[0092] ;

[0093] In the formula: , Let be the amplitude of the peak values ​​of the p-th and (p-1)-th pulses; , These are the times when the peak values ​​of the p-th and (p-1)-th pulses occur; To find the minimum value, take ;

[0094] Among them, ≥ The gradient values ​​are included in the feature dataset. This indicates a preset threshold that is dynamically adjusted based on the historical gradient mean.

[0095] The above formula calculates the ratio of the amplitude difference to the time difference of the peak values ​​of continuous pulses and introduces a minimum value to avoid calculation anomalies. Combined with a threshold dynamically adjusted based on the historical gradient mean to screen effective gradient values, it quantifies the amplitude change rate characteristics of partial discharge pulses and enhances the ability to capture the dynamic change law of pulses.

[0096] The steps for extracting the duration spectrum include:

[0097] With sliding time window , The window width is represented by 1.5 times the average pulse period. The preprocessed signal is truncated using a sliding method, and the pulse duration within the window is calculated. ;

[0098] Constructing the duration spectrum :

[0099] ;

[0100] extract The spectral peak positions and corresponding amplitudes are used as characteristic parameters;

[0101] In the formula: This is the time-domain sequence of the preprocessed partial discharge signal; For the m-th sliding time window, This represents the total number of sliding time windows; The Dirac delta function only when The value is infinity when it is infinity, and 0 in other cases, and satisfies the following conditions: ; This is a reference value for the typical duration of a partial discharge pulse; It is a time variable; The horizontal axis variable of the duration spectrum represents a general variable indicating the duration of the partial discharge pulse from start to end, covering all possible pulse duration values.

[0102] The above formula uses a window width of 1.5 times the average pulse period to slide and truncate the preprocessed signal. It uses the Dirac function to associate the actual pulse duration with a typical reference value. By constructing a spectrum, it extracts the spectral peak position and amplitude, thereby realizing a multi-scale characterization of the temporal distribution characteristics of the partial discharge pulse and improving the recognition of temporal dimension features.

[0103] The process of extracting waveform similarity includes:

[0104] The pulse waveforms with the highest amplitude in the preprocessed signal were selected as candidate templates, and the baseline template was obtained by screening using cross-correlation coefficients. ;

[0105] For any waveform to be compared After time alignment, similarity is calculated. ;

[0106] ;

[0107] In the formula: The duration of the waveform; For weighted functions, t is a time variable, representing consecutive moments on the time axis;

[0108] Will Waveforms with a similar value ≥0.75 are marked as similar waveforms, and the similarity ratio is statistically analyzed and included in the feature dataset;

[0109] The above formula selects high-amplitude pulse waveforms as candidate templates. After the benchmark template is determined by screening with cross-correlation coefficients, a weighted function is introduced to calculate the similarity of time-aligned waveforms to be compared. The statistical proportion of highly similar waveforms is included in the feature set to highlight the consistency of waveform morphology and improve the distinguishability of partial discharge signal waveform features.

[0110] A dynamic benchmark comparison model is created to perform deviation quantification analysis between the multidimensional partial discharge feature dataset and the dynamic benchmark, thereby identifying the activity level and type of partial discharge signals.

[0111] The creation of dynamic benchmark comparison models follows the following rules:

[0112] A baseline feature matrix is ​​constructed based on the most recent N normal state samples. Calculation benchmark center and feature weight matrix ;

[0113] The deviation D between the current eigenvector X and the benchmark is ;

[0114] The above formula constructs a benchmark feature matrix based on normal state samples, quantifies the degree of deviation by operating on the current feature vector, the benchmark center, and the feature weight matrix, dynamically updates the benchmark by combining a sliding window, and classifies the level of partial discharge activity based on the deviation threshold, thereby achieving accurate quantification and dynamic tracking of the degree of deviation from the normal state.

[0115] The new benchmark is updated using a sliding window. ;

[0116] The above formula constructs a benchmark feature matrix based on normal state samples, quantifies the degree of deviation by operating on the current feature vector, the benchmark center, and the feature weight matrix, dynamically updates the benchmark by combining a sliding window, and classifies the level of partial discharge activity based on the deviation threshold, thereby achieving accurate quantification and dynamic tracking of the degree of deviation from the normal state.

[0117] When D≥ At that time, according to D and The comparison is used to classify the localized activity level;

[0118] In the formula: This is the multidimensional feature vector of the Nth normal state sample; This is a transpose operation for a matrix or vector. Let be the standard deviation of the k-th feature. This is the reference center vector before the update, i.e., the reference center of the previous period; The preset deviation threshold;

[0119] Based on the time difference and waveform feature matching degree of the identified partial discharge signal arriving at each monitoring node, the spatial coordinates of the partial discharge source are obtained through distributed spatiotemporal correlation calculation.

[0120] The process of obtaining the spatial coordinates of partial discharge sources through distributed spatiotemporal correlation calculation includes:

[0121] Let the coordinates of the M monitoring nodes be... The signal arrival time is Define the source coordinates (x, y, z) of the partial discharge source to satisfy:

[0122] ;

[0123] In the formula: For signal propagation speed; This refers to the moment when partial discharge occurs; This is a correction term for propagation distance;

[0124] The spatial coordinates of the partial discharge source are obtained by solving the system of equations using the least squares method.

[0125] The above equations are combined with correction terms for signal propagation speed, occurrence time and propagation distance to construct a system of equations. The environmental attenuation coefficient is dynamically adjusted according to the density of metal obstacles and relative humidity of the air in the monitoring environment. The coordinate parameters are solved by the least squares method to achieve accurate spatial positioning of the partial discharge source in complex environments.

[0126] in, , This represents the environmental attenuation coefficient, ranging from 0.01 to 0.05. It is applied when the density of metal obstacles ρ in the monitored environment is ≥0.3 objects / m³ or the relative humidity H is ≥75%. Take a value of 0.03~0.05; when the density of metal obstacles ρ < 0.2 objects / m³ and the relative humidity H < 60%, Take 0.01~0.02; under other environmental conditions Take a value of 0.02~0.03; This represents the straight-line distance from the node to the source.

[0127] Based on the historical rate of change of partial discharge characteristics and the current multidimensional partial discharge characteristic dataset, a development trend prediction model is constructed, and based on the model, hierarchical early warning information including location, intensity and risk level is output synchronously.

[0128] The development trend prediction model is as follows:

[0129] Based on a historical partial discharge feature dataset spanning T periods, the feature growth rate is calculated. ;

[0130] Constructing the risk trend index R:

[0131] ;

[0132] In the formula: Let be the multidimensional partial discharge eigenvector of the T-th period; This is the timestamp for the Tth period; These are the weighting coefficients; This represents the deviation between the current feature vector X and the baseline. It is the average of the absolute values ​​of the differences between all adjacent periodic eigenvectors;

[0133] Specifically, when R < R1, the output is low risk; when R1 ≤ R < R2, the output is medium risk; and when R ≥ R2, the output is high risk. R1 and R2 are preset thresholds. The contribution ratios of the characteristic growth rate, deviation value, and historical volatility mean in the risk trend index are adjusted, initially set to 0.3, 0.5, and 0.2, respectively, and their sum is always 1.

[0134] The above formula integrates the historical cycle characteristic growth rate, current characteristic deviation and historical fluctuation mean, adjusts the contribution ratio of each factor through weight coefficients, classifies low, medium and high risk levels according to preset thresholds, and predicts the development trend of partial discharge by comprehensively considering multi-dimensional indicators, so as to achieve a forward-looking assessment and graded early warning of equipment partial discharge risk.

[0135] The graded early warning information also includes a partial discharge type feature matching score, which is calculated as follows:

[0136] The extracted multidimensional partial discharge feature dataset is compared with a preset partial discharge type feature library in multiple dimensions, and the type matching score is calculated. :

[0137] ;

[0138] In the formula: For feature dimensions; The type weight of the current i-th feature; This represents the current i-th eigenvalue; Let i be the standard value of the i-th feature of the t-th type of partial discharge; , Let be the maximum and minimum values ​​of the normal fluctuations of the i-th characteristic of the t-th type of partial discharge; To find the minimum value, take ;

[0139] Among them, the early warning information simultaneously outputs the partial discharge type with the highest matching score and its corresponding... numerical value, when A value ≥0.65 indicates a valid type match.

[0140] The above formula calculates the matching score by comparing the multidimensional features with the standard values ​​of the preset type feature library, combining the normal fluctuation range of the features and the feature weights determined based on the degree of dispersion, and outputs the highest matching type and score. When the score reaches the standard, the type matching is determined to be valid, highlighting the role of highly discriminative features in type recognition and improving the accuracy of partial discharge type recognition.

[0141] The value of is determined based on the degree of dispersion of the feature among different local discharge types. The more dispersed the value is among different types, the larger the value will be, and vice versa.

[0142] In this embodiment, the above method can adapt to the environment and switch capture frequency bands in real time. After multi-scale preprocessing to suppress noise and retain partial discharge pulse characteristics, it can accurately identify the type and activity of partial discharge through multi-dimensional feature extraction and dynamic benchmark comparison. Combined with distributed spatiotemporal computing, it can accurately locate the partial discharge source. Then, based on historical data, it can build a trend model and output a graded early warning that includes location, intensity, risk level and type matching. This can effectively improve the accuracy and timeliness of partial discharge monitoring and help prevent equipment failure risks in advance.

[0143] The following is an application example of the method described in the above embodiments:

[0144] To ensure the safe operation of GIS equipment, a 220kV substation adopted a partial discharge online monitoring method based on UHF raw signals for real-time monitoring. The specific application process is as follows:

[0145] Autonomous switching of UHF signal acquisition frequency bands:

[0146] The monitoring system collects broadband electromagnetic spectrum data within the substation in real time and analyzes multiple preset UHF candidate frequency bands. Calculations show that within the 1.5GHz-2.0GHz band, signal energy accounts for 0.7%, while noise energy accounts for 0.2%. Simultaneously, the ratio of the standard deviation to the mean of energy in this band is 0.12, less than 0.15; therefore, the frequency band stability factor is set to 1. Monitoring over two consecutive sampling periods shows that the signal energy proportion in this band remains above 0.6. The system then autonomously switches the UHF signal acquisition band to 1.5GHz-2.0GHz to obtain the original signal sequence covering potential partial discharge characteristics.

[0147] Multi-scale pulse contour extraction preprocessing:

[0148] The acquired original signal sequence was decomposed using variational mode decomposition (VMD) to obtain five modal components. The pulse significance index of each mode was calculated. The indexes for three modal components were 4.2, 3.8, and 5.1, respectively, all exceeding a preset threshold of 3.5. The other two modal components, whose indices were below the threshold, were discarded. Adaptive threshold denoising was then applied to the three remaining modal components. The threshold for each component was determined to be 1.4826 times the median absolute value of its signal. The reconstructed preprocessed signal after denoising effectively suppressed environmental noise interference while preserving the pulse characteristics of the partial discharge signal.

[0149] Extracting multidimensional partial discharge feature parameters:

[0150] Pulse Peak Gradient: Peak detection was performed on the preprocessed signal, detecting a total of 10 pulse peaks. When calculating the gradient between adjacent peaks, the 3rd and 2nd peaks were selected as examples, with amplitudes of 8mV and 5mV respectively, and occurrence times of 12ms and 10ms respectively. The calculated gradient between these adjacent peaks is approximately 1.5 (the minimum value is taken as 10). (The impact on the results is negligible). This gradient value is greater than the preset threshold of 1.2, which is dynamically adjusted based on the historical gradient mean, and therefore it is included in the feature dataset.

[0151] Duration spectrum: The preprocessed signal was truncated using a sliding time window width of 1.5 times the average pulse period, resulting in 20 sliding time windows. The duration spectrum was constructed by calculating the pulse duration within each window, and the peak position was extracted at 200 ns, corresponding to an amplitude of 0.8. These two parameters were included in the feature dataset.

[0152] Waveform similarity: Three pulse waveforms with the highest amplitude in the preprocessed signal were selected as candidate templates, and one of them was selected as the baseline template by screening through cross-correlation coefficient. After time alignment of all waveforms to be compared, the similarity was calculated. Statistical analysis showed that waveforms with a similarity ≥ 0.75 accounted for 80%, and this similarity percentage was included in the feature dataset.

[0153] Dynamic benchmark comparison identifies partial discharge status:

[0154] A baseline feature matrix is ​​constructed based on the most recent 50 normal state samples of the equipment, and the baseline center and feature weight matrix are calculated. The deviation quantification analysis of the currently extracted multidimensional partial discharge feature vector and the baseline is performed, and the result shows a deviation value of 1.2, which is greater than the preset deviation threshold of 0.8. Based on the comparison of the deviation value with the baseline, the current partial discharge signal activity level is determined to be medium. Simultaneously, the system uses a sliding window to update the baseline, ensuring that the baseline can adapt to the slow changes in equipment status.

[0155] Location of localized source in distributed spatiotemporal correlation computation:

[0156] Three monitoring nodes are set up within the substation, with coordinates of (5m, 3m, 2m), (20m, 5m, 2m), and (10m, 15m, 2m). Based on the time difference of the partial discharge signal arriving at each node and the waveform characteristic matching degree, a system of equations is constructed to solve for the coordinates of the partial discharge source. Considering the density of metal obstacles in the monitoring environment is 0.25 objects / m³, the relative humidity is 65%, and the environmental attenuation coefficient is taken as 0.025, the equations are solved using the least squares method, ultimately yielding the spatial coordinates of the partial discharge source as (15m, 8m, 3m).

[0157] Development Trend Forecast and Tiered Early Warning:

[0158] Based on a historical partial discharge (PD) feature dataset spanning 10 cycles, the feature growth rate was calculated. Combining the deviation of the current multidimensional PD feature vector from the baseline (1.2) and the average absolute value of the differences between feature vectors from all adjacent cycles (0.3), a risk trend index was constructed. The weighting coefficients were set to 0.3, 0.5, and 0.2, resulting in a calculated risk trend index of 0.6. Since the preset low-risk threshold R1 = 0.4 and the medium-risk threshold R2 = 0.7, the current index falls within the range of R1 ≤ R < R2, thus classifying it as medium risk. Simultaneously, the multidimensional PD feature dataset was compared with a preset PD type feature library, and a type matching score was calculated. The matching score for suspended potential discharge was 0.72, greater than 0.65, indicating a valid type match. Finally, the system synchronously outputs graded early warning information including the PD source location (15m, 8m, 3m), intensity (medium), and risk level (medium risk), labeling the PD type as suspended potential discharge.

[0159] In summary, the methods described in the above embodiments, during execution, autonomously switch UHF acquisition bands through real-time electromagnetic spectrum analysis, dynamically adapt to changes in environmental noise, and improve the accuracy of original signal acquisition. Multi-scale pulse profile extraction preprocessing effectively suppresses noise interference while preserving key pulse features. A multi-dimensional feature dataset constructed based on pulse peak gradient, duration spectrum, and waveform similarity provides comprehensive evidence for partial discharge identification. The dynamic benchmark comparison model accurately identifies the activity level and type of partial discharge through deviation quantification. Distributed spatiotemporal correlation calculation combined with propagation correction terms achieves precise localization of partial discharge sources. The development trend prediction model integrates historical feature changes with current data, outputting a graded early warning including location, intensity, risk level, and type matching scores. The weighting coefficient dynamically adjusts the risk contribution ratio to ensure the scientific nature of the early warning. Overall, this approach optimizes the entire process of partial discharge monitoring, from signal acquisition and feature extraction to identification, localization, and risk warning, improving monitoring reliability and early warning capabilities.

[0160] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A partial discharge online monitoring method based on ultra-high frequency raw signals, characterized in that, include: Based on the real-time electromagnetic spectrum distribution of the monitoring environment, the system autonomously switches the UHF signal acquisition frequency band to obtain the original signal sequence covering potential partial discharge characteristics. The original signal sequence is preprocessed by multi-scale pulse profile extraction to suppress environmental noise interference and preserve the pulse characteristics of the partial discharge signal. Three spatiotemporal distribution parameters—pulse peak gradient, duration spectrum, and waveform similarity—are extracted from the preprocessed signal to construct a multidimensional partial discharge feature dataset. A dynamic benchmark comparison model is created to perform deviation quantification analysis between the multidimensional partial discharge feature dataset and the dynamic benchmark, thereby identifying the activity level and type of partial discharge signals. Based on the time difference and waveform feature matching degree of the identified partial discharge signal arriving at each monitoring node, the spatial coordinates of the partial discharge source are obtained through distributed spatiotemporal correlation calculation. Based on the historical rate of change of partial discharge characteristics and the current multidimensional partial discharge characteristic dataset, a development trend prediction model is constructed, and based on the model, hierarchical early warning information including location, intensity and risk level is output synchronously.

2. The partial discharge online monitoring method based on UHF raw signals according to claim 1, characterized in that, The specific process of autonomously switching the UHF signal acquisition frequency band based on the real-time electromagnetic spectrum distribution of the monitoring environment includes: Real-time acquisition of broadband electromagnetic spectrum data of the monitoring environment, and calculation of the noise energy proportion of each preset ultra-high frequency sub-band. and signal energy ratio And determine the optimal acquisition frequency band based on the following formula; ； In the formula: This is the optimal ultra-high frequency signal capture band; This is a pre-defined set of UHF candidate frequency bands; This represents the percentage of signal energy within frequency band f at time t. This represents the percentage of noise energy in frequency band f at time t. This is the frequency band stability factor, with a value range of [0,1]. in, The value follows this rule: when the ratio of the standard deviation to the mean of the frequency band energy is less than or equal to 0.

15. =1; when the ratio of the standard deviation to the mean of the frequency band energy is greater than 0.15, The frequency band decreases linearly as the ratio increases, and frequency band switching is performed when the signal energy ratio of the optimal frequency band for two consecutive sampling periods is not less than 0.

6.

3. The partial discharge online monitoring method based on UHF raw signals according to claim 1, characterized in that, The multi-scale pulse contour extraction preprocessing operation of the original signal sequence follows: Variational mode decomposition is used to decompose the original signal sequence x(t) into K modal components. ; Calculate the impulse significance index for each mode. ; ； In the formula: The absolute maximum value of the k-th modal component signal; The mean of the absolute values ​​of the k-th modal component signal; Let V be the variance of the absolute value of the k-th modal component signal; Among them, retain ≥ The modal components are analyzed, and adaptive threshold denoising is applied to the retained components to reconstruct the preprocessed signal. The preset threshold is set to 3.5; the threshold value is used in the adaptive threshold denoising process. obey: , This represents the median of the absolute values ​​of the k-th modal component signal.

4. The partial discharge online monitoring method based on UHF raw signals according to claim 1, characterized in that, The operation of extracting the pulse peak gradient includes: Peak detection is performed on the preprocessed signal to obtain the peak amplitude sequence. and corresponding time , This indicates the total number of detected pulse peaks; Define the gradient of adjacent peaks : ； In the formula: , Let be the amplitude of the peak values ​​of the p-th and (p-1)-th pulses; , These are the times when the peak values ​​of the p-th and (p-1)-th pulses occur; To find the minimum value, take ; Among them, ≥ The gradient values ​​are included in the feature dataset. This indicates a preset threshold that is dynamically adjusted based on the historical gradient mean.

5. The partial discharge online monitoring method based on UHF raw signals according to claim 1, characterized in that, The extraction steps of the duration spectrum include: With sliding time window , The window width is represented by 1.5 times the average pulse period. The preprocessed signal is truncated using a sliding method, and the pulse duration within the window is calculated. ; Constructing the duration spectrum : ； extract The spectral peak positions and corresponding amplitudes are used as characteristic parameters; In the formula: This is the time-domain sequence of the preprocessed partial discharge signal; For the m-th sliding time window, This represents the total number of sliding time windows; The Dirac delta function only when The value is infinity when it is infinity, and 0 in other cases, and satisfies the following conditions: ; This is a reference value for the typical duration of a partial discharge pulse; It is a time variable; The horizontal axis variable of the duration spectrum represents a general variable indicating the duration of the partial discharge pulse from start to end, covering all possible pulse duration values.

6. The partial discharge online monitoring method based on UHF raw signals according to claim 1, characterized in that, The process of extracting waveform similarity includes: The pulse waveforms with the highest amplitude in the preprocessed signal were selected as candidate templates, and the baseline template was obtained by screening using cross-correlation coefficients. ; For any waveform to be compared After time alignment, similarity is calculated. ; ； In the formula: The duration of the waveform; For weighted functions, t is a time variable, representing consecutive moments on the time axis; Will Waveforms with a value ≥0.75 are marked as similar waveforms, and the proportion of similarity is statistically analyzed and included in the feature dataset.

7. The partial discharge online monitoring method based on UHF raw signals according to claim 1, characterized in that, The creation of the dynamic benchmark comparison model follows the following: A baseline feature matrix is ​​constructed based on the most recent N normal state samples. Calculation benchmark center and feature weight matrix ; The deviation D between the current eigenvector X and the benchmark is ; The new benchmark is updated using a sliding window. ; When D≥ At that time, according to D and The comparison is used to classify the localized activity level; In the formula: This is the multidimensional feature vector of the Nth normal state sample; This is a transpose operation for a matrix or vector. Let be the standard deviation of the k-th feature. This is the reference center vector before the update, i.e., the reference center of the previous period; This is the preset deviation threshold.

8. The partial discharge online monitoring method based on UHF raw signals according to claim 1, characterized in that, The process for obtaining the spatial coordinates of the partial discharge source through distributed spatiotemporal correlation calculation includes: Let the coordinates of the M monitoring nodes be... The signal arrival time is Define the source coordinates (x, y, z) of the partial discharge source to satisfy: ； In the formula: For signal propagation speed; This refers to the moment when partial discharge occurs; This is a correction term for propagation distance; The spatial coordinates of the partial discharge source are obtained by solving the system of equations using the least squares method. in, , This represents the environmental attenuation coefficient, ranging from 0.01 to 0.

05. It is applied when the density of metal obstacles ρ in the monitored environment is ≥0.3 objects / m³ or the relative humidity H is ≥75%. Take a value of 0.03~0.05; when the density of metal obstacles ρ < 0.2 objects / m³ and the relative humidity H < 60%, Take 0.01~0.02; under other environmental conditions Take a value of 0.02~0.03; This represents the straight-line distance from the node to the source.

9. The partial discharge online monitoring method based on UHF raw signals according to claim 1, characterized in that, The development trend prediction model is as follows: Based on a historical partial discharge feature dataset spanning T periods, the feature growth rate is calculated. ; Constructing the risk trend index R: ； In the formula: Let be the multidimensional partial discharge eigenvector of the T-th period; This is the timestamp for the Tth period; These are the weighting coefficients; This represents the deviation between the current feature vector X and the baseline. It is the average of the absolute values ​​of the differences between all adjacent periodic eigenvectors; Specifically, when R < R1, the output is low risk; when R1 ≤ R < R2, the output is medium risk; and when R ≥ R2, the output is high risk. R1 and R2 are preset thresholds. The contribution ratios of the characteristic growth rate, deviation value, and historical volatility mean in the risk trend index are adjusted, initially set to 0.3, 0.5, and 0.2, respectively, and their sum is always 1.

10. The partial discharge online monitoring method based on UHF raw signals according to claim 1, characterized in that, The graded early warning information also includes a partial discharge type feature matching score, the calculation process of which is as follows: The extracted multidimensional partial discharge feature dataset is compared with a preset partial discharge type feature library in multiple dimensions, and the type matching score is calculated. : ； In the formula: For feature dimensions; The type weight of the current i-th feature; This represents the current i-th eigenvalue; Let i be the standard value of the i-th feature of the t-th type of partial discharge; , Let be the maximum and minimum values ​​of the normal fluctuations of the i-th characteristic of the t-th type of partial discharge; To find the minimum value, take ; Among them, the early warning information simultaneously outputs the partial discharge type with the highest matching score and its corresponding... numerical value, when A value ≥0.65 indicates a valid type match. The value of is determined based on the degree of dispersion of the feature among different local discharge types. The more dispersed the value is among different types, the larger the value will be, and vice versa.

Images