A voltage transformer discharge detection method, system and device

Abstract

The application discloses a voltage transformer discharge detection method, system and equipment, and relates to the technical field of power equipment detection, which comprises the following steps: determining a local high-risk insulation sector of a voltage transformer and establishing a radial reference field, collecting multi-source data through a sensor and performing calibration excitation modeling; converting a discrete event observation model into a frequency domain representation according to the multi-source data, extracting an event window according to a threshold of the sensor, and adopting a multi-candidate algorithm to reconstruct a result and calculate a comprehensive score to select an optimal reconstruction result; determining a single event equivalent radial source position according to the optimal reconstruction result, calculating a propagation distance to perform geometric correction, constructing a propagation transfer function to compensate for the geometric correction to obtain a channel standardized waveform; and the application realizes robust recovery of a discharge event under a complex noise background, and improves the accuracy, stability and engineering applicability of voltage transformer defect classification.

Description

Technical Field

[0001] This invention relates to the field of power equipment testing technology, and in particular to a method, system and equipment for detecting discharge of voltage transformers. Background Technology

[0002] With the development of power systems towards higher voltage levels, digital monitoring, and condition-based maintenance, voltage transformers, as crucial primary equipment for energy metering, relay protection, condition sensing, and fault analysis, have seen their insulation health status become a key factor affecting the safe operation of substations and equipment lifespan management. Partial discharge, as a significant precursor signal of insulation degradation, interface defects, moisture aging, and internal air gap instability, exhibits early-onset, concealed, and evolving characteristics, making it a key monitoring target in voltage transformer fault early warning. Existing voltage transformer discharge detection technologies have generally evolved from single-signal detection to multi-source collaborative sensing, from empirical threshold discrimination to feature modeling identification, and from offline experimental analysis to online condition assessment. However, existing technologies still have limitations. Many solutions rely primarily on single-channel or low-level features for discrimination, making it difficult to consider the coupling characteristics of partial discharge within the voltage transformer in terms of phase, space, propagation path, and dielectric response. This results in insufficient accuracy in identifying weak discharges, composite defects, and early-evolutionary anomalies. Summary of the Invention

[0003] In view of the aforementioned existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides a method, system and device for detecting discharge in voltage transformers, which solves the problem that the existing technology is unable to take into account the coupling characteristics of partial discharge inside voltage transformers in terms of phase, space, propagation path and medium response.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] In a first aspect, the present invention provides a method for detecting discharge in a voltage transformer, comprising,

[0007] Identify local high-risk insulation sectors of voltage transformers and establish radial reference fields; collect multi-source data through sensors and perform calibration excitation modeling.

[0008] A discrete event observation model is established based on multi-source data and converted into a frequency domain representation. A threshold is established based on sensors to extract events and output event windows. For each event window, a multi-candidate algorithm is used to reconstruct the results and a comprehensive score is calculated to select the optimal reconstruction result.

[0009] Based on the optimal reconstruction results, the equivalent radial source position of a single event is determined and the propagation distance is calculated for geometric correction. The propagation transfer function is constructed to compensate for the geometric correction and obtain the channel-normalized waveform.

[0010] The channel-standardized waveforms are normalized to construct the main cluster waveform signature of the events. The clustering distance is calculated to perform clustering. The events are mapped to the phase window index and spatial index. The weighted event set is constructed by combining the clustering results.

[0011] The total intensity of the phase cluster and the spatial concentration factor within the cluster are defined based on the weighted event set, and the joint fingerprint unit and the total joint coupling strength are also defined.

[0012] A multi-dimensional feature vector is established based on the joint fingerprint unit and the total joint coupling strength and then standardized. Based on the standardized multi-dimensional feature vector, the defect category of the voltage transformer is identified.

[0013] As a preferred embodiment of the voltage transformer discharge detection method of the present invention, the step of determining the local high-risk insulation sector of the voltage transformer and establishing a radial reference field, and acquiring multi-source data through sensors and performing calibration excitation modeling refers to obtaining voltage transformer structural data to select the local high-risk insulation sector.

[0014] In the insulated mountain area, the average outer radius of the outer high-voltage winding is defined as the equivalent inner radius. And define the equivalent outer radius ;

[0015] Define local equivalent peak voltage And establish a radial reference field for a locally axisymmetric equivalent insulation channel. ;

[0016] Multi-source sensors are deployed on the voltage transformer to collect multi-source data, including 1 HFCT sensor, 2 UHF sensors, 4 ultrasonic sensors, 2 temperature sensors, and 1 power frequency reference voltage sampling terminal.

[0017] In local high-risk insulation sectors The calibration excitation model was modeled using an electrical calibration excitation actuator and a piezoelectric acoustic calibration actuator.

[0018] As a preferred embodiment of the voltage transformer discharge detection method of the present invention, the steps of establishing a discrete event observation model based on multi-source data and converting it into a frequency domain representation, establishing a threshold to extract event output windows based on sensors, and using a multi-candidate algorithm to reconstruct the results for each event window and calculating a comprehensive score to select the optimal reconstruction result include:

[0019] A discrete event observation model is established based on the collected multi-source data, and the discrete event observation model is converted into a frequency domain representation.

[0020] The number of sampling points is defined for the HFCT sensor channel. The front window is used to calculate the mean within the front window. and standard deviation HFCT trigger thresholds were established based on the mean and standard deviation. ;

[0021] When HFCT sensor channel data When the time is greater than or equal to the trigger threshold, Defined as the main trigger time of the i-th candidate event, and based on time... Build Event Window ;

[0022] The validity of candidate events is further determined based on the UHF and ultrasonic sensors, and an event window is output. ;

[0023] The result is reconstructed using a multi-candidate algorithm for the i-th event window;

[0024] The scoring criteria, including the reconvolution residual ratio, are calculated based on the reconstruction results of the multi-candidate algorithm. Baseline fluctuation ratio and peak concentration penalty ;

[0025] The comprehensive score for the reconstruction results of each candidate algorithm is obtained by weighted summation of the scoring items. The candidate algorithm with the highest comprehensive score is selected as the optimal reconstruction result. .

[0026] As a preferred embodiment of the voltage transformer discharge detection method of the present invention, the step of determining the equivalent radial source position of a single event based on the optimal reconstruction result and calculating the propagation distance for geometric correction, and constructing a propagation transfer function to compensate for the geometric correction to obtain the channel-normalized waveform includes:

[0027] The equivalent radial source location for a single event was determined based on the optimal ultrasound reconstruction results. ;

[0028] The propagation distance of the i-th event to the UHF and ultrasonic sensor is calculated based on the equivalent radial source location;

[0029] Define a position excitation correction factor and a geometric diffusion compensation factor for the i-th event;

[0030] Geometric correction of UHF and ultrasonic sensor channels;

[0031] By repeatedly injecting into the piezoelectric calibration actuator The same stimulus was used again, and the calibration response was collected. And on The calibration frequency domain is obtained by averaging the responses synchronously and then performing a Fourier transform. The ultrasonic sensor channel with the shortest propagation distance was selected as the reference channel, and the reference calibration frequency domain was obtained. Define the relative propagation transfer function for any non-reference channel. ;

[0032] By propagating the transfer function Obtain the parameters of the j-th ultrasound channel;

[0033] The parameters of all ultrasound channels are weighted and synthesized into a common parameter. Based on the common parameter, the propagation transfer function of the UHF and ultrasound sensor channels is formed. And compensate for the geometric correction results of the UHF and ultrasonic sensor channels;

[0034] The compensated channel normalized waveform is obtained by using inverse Fourier transform. ;

[0035] Calculate the area response of each channel for the i-th event. ;

[0036] The mean comprehensive scores for HFCT, UHF, and ultrasound channels are calculated based on the maximum comprehensive score for each channel. Further, an exponential function is used to calculate the reliability weights for HFCT, UHF, and ultrasound channels respectively. These reliability weights are then normalized. Based on these normalized reliability weights, the area-type response quantities are weighted and fused to obtain the fused response quantity for the i-th event. ;

[0037] The fused response is mapped to an uncalibrated equivalent response curve distributed along a local radial direction. ;

[0038] Construct a reference equivalent response curve And calculate the uniform calibration coefficient. ;

[0039] The equivalent response charge of the i-th event after calibration is obtained based on the unified calibration coefficient. and equivalent response density ;

[0040] For the i-th event, a radial first-order differential equation is established, and the equivalent insulation stress response distribution is obtained by solving the radial first-order differential equation. ;

[0041] exist to Within the range, the largest equivalent insulation stress response distribution is selected as the peak insulation stress for the i-th event. The enhancement factor is calculated based on the peak insulation stress. .

[0042] As a preferred embodiment of the voltage transformer discharge detection method of the present invention, the following steps are performed: Normalizing the channel standardized waveforms to construct the main cluster waveform signature of the event; calculating the clustering distance; performing clustering index; performing energy normalization on the standardized waveforms of the two UHF sensor channels for the i-th event; exponentially calculating the fusion weight of the two UHF sensor channels by exponentializing the maximum comprehensive score of each UHF sensor channel; and constructing the main cluster waveform signature of the i-th event based on the fusion weight. ;

[0043] Define window For windows For any two events i and j within the range, calculate the elements of the globally normalized cross-correlation matrix. ;

[0044] The elements of the globally normalized cross-correlation matrix are used to form a similarity matrix. And convert it into the main distance matrix. ;

[0045] Based on the principal distance matrix Calculate the cluster distance between events i and j ;

[0046] Cluster distances are used to form a distance matrix, and average connectivity hierarchical clustering is used for clustering. The clustering stopping condition is set to the point where the average distance between any two clusters is greater than a threshold. Stop merging when the clustering is complete and output the clustering results.

[0047] As a preferred embodiment of the voltage transformer discharge detection method of the present invention, the step of mapping events to phase window index and spatial index, and constructing a weighted event set based on clustering results to calculate the window... The power frequency phase of the i-th event ;

[0048] Based on the power frequency phase, 12 equal-width phase windows are used to map the i-th event to the phase window index. ;

[0049] Local equivalent insulation channel The space is fixedly divided into four radial regions of equal width, and the spatial index of the i-th event is defined. ;

[0050] Based on phase window index Spatial Index And construct a weighted event set from the clustering results. .

[0051] In a preferred embodiment of the voltage transformer discharge detection method of the present invention, the following steps are taken: defining the total intensity of the phase cluster and the spatial concentration factor within the cluster based on the weighted event set; defining the joint fingerprint unit and the joint coupling total intensity; calculating the dual UHF average score of the i-th event; and defining the physical consistency factor of the i-th event relative to the mean of its cluster. ;

[0052] Event credibility weights are calculated based on physical consistency factors and average scores from both UHF systems. And define the equivalent discharge amplitude. ;

[0053] For weighted event set Each group Define weighted discharge intensity ;

[0054] The total intensity of the phase cluster is obtained by summing the weighted discharge intensities in the radial spatial region for the same phase window and cluster. And calculate the spatial concentration factor within the cluster. ;

[0055] Simultaneous calculation of weighted average peak stress for the same phase window and cluster ;

[0056] Calculate the local equivalent temperature based on the temperature sensor. Constructing phase-cluster stress enhancement factor ;

[0057] Define a joint fingerprint unit for each phase window p and cluster e. ;

[0058] The total joint coupling strength is obtained by summing the joint fingerprint units. .

[0059] As a preferred embodiment of the voltage transformer discharge detection method of the present invention, the step of establishing a multi-dimensional feature vector based on the joint fingerprint unit and the joint total coupling strength and then standardizing it, and identifying the voltage transformer defect category based on the standardized multi-dimensional feature vector, includes:

[0060] Define multidimensional feature vectors Median-interquartile range robust normalization is used for multidimensional feature vectors. The process yields a standardized multidimensional feature vector. ;

[0061] A six-category defect system is defined, and for each defect category, a linear discriminant score is used. ;

[0062] Linear discriminant score The probability of each defect category is obtained by using softmax normalization, and the defect category with the highest probability is selected as the final defect.

[0063] Secondly, the present invention provides a voltage transformer discharge detection system, comprising,

[0064] The data acquisition module is used to identify local high-risk insulation sectors of the voltage transformer and establish a radial reference field. It collects multi-source data through sensors and performs calibration excitation modeling.

[0065] The event extraction module is used to build a discrete event observation model based on multi-source data and convert it into a frequency domain representation. It also builds a threshold to extract events based on sensors and outputs event windows. For each event window, it uses a multi-candidate algorithm to reconstruct the results and calculates a comprehensive score to select the optimal reconstruction result.

[0066] The geometry correction module is used to determine the equivalent radial source position of a single event based on the optimal reconstruction result and calculate the propagation distance for geometry correction. It also constructs a propagation transfer function to compensate for the geometry correction and obtain the channel-normalized waveform.

[0067] The clustering module is used to normalize the channel-standardized waveforms, construct the master cluster waveform signature of events, calculate the cluster distance, perform clustering, map events to phase window indices and spatial indices, and construct a weighted event set by combining the clustering results.

[0068] The defect analysis module is used to define the total intensity of phase clusters and the spatial concentration factor within the clusters based on the weighted event set, and to define the joint fingerprint unit and the total joint coupling intensity. It establishes a multi-dimensional feature vector based on the joint fingerprint unit and the total joint coupling intensity for standardization, and identifies the defect category of the voltage transformer based on the standardized multi-dimensional feature vector.

[0069] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the voltage transformer discharge detection method as described in the first aspect of the present invention.

[0070] The beneficial effects of this invention are as follows: By identifying local high-risk insulation sectors of voltage transformers and establishing a radial reference field, combined with multi-source sensing acquisition and calibration excitation modeling, this invention transforms the discharge observation range from coarse monitoring to directional constraint of local insulation sensitive areas, thereby improving the detection targeting from the source; by establishing a discrete event observation model, event window extraction, and multi-candidate reconstruction optimization, robust recovery of discharge events under complex noise backgrounds is achieved, providing reliable input for subsequent analysis; through single-event equivalent radial source localization, propagation distance geometric correction, and propagation compensation, channel-standardized waveforms with stronger comparability are obtained, which helps to reduce the influence of path differences and sensing differences; furthermore, through clustering, phase-space mapping, and weighted event set construction, the discharge activity is improved from single-point features to cluster behavior features; finally, by combining joint fingerprint units and joint coupling total strength, multi-dimensional defect identification is completed, thereby improving the accuracy, stability, and engineering applicability of voltage transformer defect classification. Attached Figure Description

[0071] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0072] Figure 1 This is a flowchart of the voltage transformer discharge detection method in Example 1.

[0073] Figure 2 This is a structural diagram of the voltage transformer discharge detection method in Example 1. Detailed Implementation

[0074] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0075] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0076] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0077] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides a method for detecting discharge in a voltage transformer, comprising the following steps:

[0078] S1. Identify the local high-risk insulation sector of the voltage transformer and establish a radial reference field. Collect multi-source data through sensors and perform calibration excitation modeling.

[0079] S1.1 Obtaining voltage transformer structural data: Selecting local high-risk insulation sectors, specifically, in the radial section of the voltage transformer, along the height position of the winding midpoint. Select the outer surface of the outer high-voltage winding, where H is the axial height of the winding, and start from the middle height position. The nearest grounding shield or metal casing surface is located along the normal direction, and the shortest primary insulation path between the two is defined as the unique local high-risk insulation sector. ;

[0080] Within the insulation sector, the average outer radius of the outer high-voltage winding is defined as the equivalent inner radius. And define the equivalent outer radius :

[0081]

[0082] in This refers to the normal distance of the main insulation between the outer surface of the high-voltage winding and the nearest grounded shielding layer. Represents the inner boundary radius of the locally equivalent insulation channel. Represents the outer boundary radius of the locally equivalent insulation channel;

[0083] Define local equivalent peak voltage And establish a radial reference field for a locally axisymmetric equivalent insulation channel. :

[0084]

[0085]

[0086] in The effective value of the primary relative voltage to ground is determined based on the voltage transformer parameters. This is the local voltage distribution coefficient. For the equivalent center radius, Let r be the local reference electric field at a distance r from the equivalent center. The maximum reference field strength at the inner boundary;

[0087] It should be noted that the local voltage distribution coefficient The value ranges from 0.90 to 1.00, with a preferred value of 0.95. In a real voltage transformer, the electric field is not applied uniformly within the local insulation sector, and there are edge shielding and winding distribution effects. If the value is directly set to 1, the local stress will be overestimated. If the value is too low, the subsequent propagation and discharge triggering sensitivity will be underestimated. Therefore, a value of 0.95 is chosen to balance conservatism and feasibility.

[0088] S1.2 Deploy multi-source sensors on the voltage transformer to collect multi-source data, including 1 HFCT sensor (high frequency current transformer), 2 UHF sensors (ultra-high frequency sensors), 4 ultrasonic sensors, 2 temperature sensors, and 1 power frequency reference voltage sampling terminal.

[0089] Specifically, the sensor deployment is as follows:

[0090] First, install one HFCT sensor on a single main section of the main grounding down conductor. The axial distance between the installation position and the grounding weld point of the fuel tank is fixed at [value missing]. The preferred distance is 120mm. If the distance is too close to the solder joint, the vibration of the metal structure and the parasitic inductance of the solder joint will increase the interference. If the distance is too far, it is easy to introduce the error of the branch grounding loop. Choosing 120mm ensures that the error and interference are balanced.

[0091] Secondly, at the same horizontal cross-section of the fuel tank Install two UHF sensors, located at polar angles respectively. By using two UHF sensors to form a symmetrical observation channel, it is easy to determine the difference in electromagnetic pulse propagation direction and suppress the bias of local external interference.

[0092] Next, at the same horizontal cross section Four ultrasonic sensors are installed, with polar angles distributed at 90°:

[0093]

[0094] in For sensor serial number, Install polar angles for the ultrasonic sensor;

[0095] Two temperature sensors are deployed in different locations, with the first temperature sensor... Fixed in the upper hot oil zone, axial height is taken as The second temperature sensor Fixed in local high-risk insulation sectors Near the projection on the outer wall of the fuel tank, the axial height Temperature sensor Temperature sensor used to characterize global thermal state Used to characterize the thermal drift of locally insulated sectors;

[0096] The power frequency reference voltage sampling terminal is deployed in the secondary side stable measurement circuit;

[0097] S1.3, in localized high-risk insulation sectors Calibration excitation modeling is performed using an electrical calibration actuator and a piezoelectric acoustic calibration actuator:

[0098] In local high-risk insulation sectors An electrical calibration coupling electrode is fixedly installed at the corresponding projection position of the outer casing. This electrode is used to calibrate the HFCT and UHF electrical channels. A narrow pulse calibration signal is applied to the locally insulated channel through this electrode. The calibration pulse waveform is a unipolar Gaussian pulse.

[0099]

[0100] in For electrical calibration excitation voltage, To electrically calibrate the peak amplitude of the pulse, For time, The pulse center time, For pulse width parameters, the preferred option is... It is 2ns;

[0101] In local high-risk insulation sectors A piezoelectric actuator is installed at the corresponding projection position of the housing to calibrate the ultrasonic channel, ensuring it operates at the same horizontal cross-section as the ultrasonic sensor array. The acoustic school uses a damped sinusoidal pulse for excitation.

[0102]

[0103] in To calibrate the actuator drive voltage for acoustic calibration, To drive the amplitude, The damping coefficient is... Angular frequency;

[0104] For a sound wave generated by a piezoelectric actuator and propagating in the housing, the sound pressure wave propagating to a distance r is defined as follows. :

[0105]

[0106] in For the propagation angular frequency, The source-end sound pressure level is the amplitude. It is an imaginary number. For complex wave number, , The attenuation constant determines how much a frequency component will attenuate after propagating a certain distance. As a phase constant, it determines how far a frequency component propagates before accumulating phase lag.

[0107] Based on the definition of sound pressure wave propagation transfer function :

[0108]

[0109] in This represents the sound pressure spectrum at the source end.

[0110] S2. Based on multi-source data, establish a discrete event observation model and convert it into a frequency domain representation. Based on the sensor, establish a threshold to extract the event output window. For each event window, use a multi-candidate algorithm to reconstruct the result and calculate a comprehensive score to select the optimal reconstruction result.

[0111] S2.1 Establish a discrete event observation model based on the collected multi-source data:

[0112]

[0113] in This refers to multi-source data collected by the sensor, including the sensor's raw observed waveforms. The waveform to be recovered at the sensor input is the true response waveform. This represents the equivalent impulse response of the channel from the event source to the input terminal of the data acquisition card. For noise, c represents the sensor channel, including the HFCT channel, UHF channel, and ultrasound channel, and j represents the sensor number.

[0114] Specifically, noise item The comprehensive equivalent disturbance term of the c-th sensor channel on the j-th sensor is independent of the target event. The sources include at least sensor body noise, analog front-end noise, connection line and coupling noise, operating environment disturbance noise, and analog-to-digital conversion quantization noise. The noise term does not need to be separated independently, but is jointly estimated by the static steady segment before the event is triggered, the idle sampling segment without events, and the silent sampling segment during the calibration stage. The noise mean, variance, and power spectral density are further calculated on the noise sample set as noise prior inputs for subsequent backpropagation compensation, regularization recovery, threshold determination, and channel weighting.

[0115] Convert the discrete event observation model into a frequency domain representation:

[0116]

[0117] in , , They are respectively , , The frequency domain representation, The equivalent frequency domain transfer function of the channel;

[0118] The estimated value of the equivalent frequency domain transfer function is calculated based on the calibration excitation model:

[0119]

[0120] in Let be the estimated value of the measured equivalent transfer function of the j-th sensor in the c-th sensor channel. The response spectrum obtained during channel calibration is acquired by collecting calibration excitation response data through sensors. The input spectrum for the corresponding calibration actuator is obtained through the actuator;

[0121] It should be noted that the calibration actuator input spectrum Corresponding to the sensor channels, the HFCT and UHF electrical channels use unipolar Gaussian pulse signals. The input spectrum and response spectrum of the corresponding sensor channel are obtained through these unipolar Gaussian pulse signals. The response spectrum is formed by the response signals of the unipolar Gaussian pulse signal acquired by the sensor in the HFCT and UHF electrical channels. Based on the input spectrum and response spectrum during calibration, the equivalent transfer function estimate of the HFCT and UHF electrical channels is calculated inversely using the frequency domain representation of the discrete event observation model. The calibration noise term can be reduced to a negligible level through repeated injection and synchronous averaging. Therefore, the noise term is ignored in the first-order estimation. For the ultrasonic channel, a damped sinusoidal pulse signal is used. The input spectrum is obtained through the damped sinusoidal pulse signal, and the equivalent transfer function estimate of the ultrasonic channel is calculated by acquiring the response spectrum. The response spectrum of the ultrasonic channel is converted into an electrical signal after the ultrasonic sensor acquires the sound wave signal and the spectrum is calculated. The ultrasonic calibration adopts a fixed installation position, fixed coupling method, fixed excitation amplitude and preset temperature compensation conditions. Within the statistical window corresponding to each event analysis, the shell-oil medium-sensor composite propagation link is approximated as a short-time quasi-linear, quasi-time-invariant system. The relative propagation transfer function is used for parameter extraction and propagation compensation only when the calibration signal-to-noise ratio is higher than the threshold, the reference channel peak drift is less than the threshold, and the correlation coefficient between two adjacent calibration responses is higher than the threshold.

[0122] S2.2, The number of sampling points is defined for the HFCT sensor channel. The front window is used to calculate the mean within the front window. and standard deviation HFCT trigger thresholds were established based on the mean and standard deviation. :

[0123]

[0124] in The HFCT trigger coefficient is preferably set to 6. When the value is less than 5, power frequency harmonics, grounding inrush current, and external narrow pulses can easily cause false triggering. When the value is greater than 8, early weak partial discharge events are easily missed. Therefore, a value of 6 is optimal between sensitivity and false alarm rate.

[0125] When HFCT sensor channel data When the time is greater than or equal to the trigger threshold, Defined as the main trigger time of the i-th candidate event, and based on time... Build Event Window :

[0126]

[0127]

[0128]

[0129]

[0130] in , , These are the observation windows for HFCT, UHF, and ultrasound sensor channel data, respectively. and To observe the moment in front of the window, , and To observe the time after the window, , This refers to the time following the ultrasound channel observation window. The effective response of both HFCT and UHF channels occurs in the microsecond range, while the ultrasonic channel requires a millisecond-level window coverage due to the large propagation delay of the housing and oil medium.

[0131] The validity of candidate events is further determined based on the UHF and ultrasonic sensors, and an event window is output. :

[0132] For any UHF sensor channel and ultrasonic channel, calculate the mean and standard deviation of the UHF sensor and ultrasonic sensor data within the corresponding channel's pre-window, and calculate the UHF trigger threshold based on the mean and standard deviation. and ultrasound trigger threshold If any UHF sensor channel data satisfy And at least two ultrasound channel data satisfy If so, the event will be recognized as a valid event, and the event window will be output.

[0133] S2.3, Reconstruct the result of the i-th event window using a multi-candidate algorithm;

[0134] Specifically, the multi-candidate algorithm includes:

[0135] Candidate 1 uses Fourier direct deconvolution to calculate the frequency domain candidate recovery spectrum for the data of the j-th sensor channel in the c-th sensor channel within the i-th time window:

[0136]

[0137] in For the candidate recovery spectrum of Fourier direct deconvolution, Let be the spectrum of the data from the j-th sensor channel within the c-th sensor channel;

[0138] Performing an inverse Fourier transform on the frequency domain candidate recovery spectrum yields the candidate 1 reconstructed true response waveform. ;

[0139] Candidate 2 is a Wiener deconvolution, and the Wiener candidate recovery spectrum is constructed as follows:

[0140]

[0141] in for conjugate, Let be the Wiener regularization coefficient for the c-th sensor channel, preferably set to a value of . , , ,in The mean noise power spectrum of the c-th sensor channel in the silent region is obtained by calculating the pre-static stable region of the event window. This represents the mean noise power spectrum of the silent zone of the HFCT sensor channel. This represents the mean noise power spectrum of the UHF sensor channel in the silent region. The noise power spectrum mean of the silent zone of the ultrasonic sensor is obtained by acquiring the noise power spectrum of each sensor channel in the pre-static stable zone and averaging it. When averaging, it is necessary to consider that there is 1 channel for HFCT sensor, 2 channels for UHF sensor, and 4 channels for ultrasonic sensor.

[0142] Performing an inverse Fourier transform on the Wiener candidate recovery spectrum yields the candidate 2 reconstructed true response waveform. ;

[0143] Candidate 3 is Tikhonov temporal deconvolution, which transforms the discrete event observation model into matrix form:

[0144]

[0145] in The estimated value of the equivalent impulse response The constructed Toeplitz convolution matrix, pass Obtained by inverse transformation;

[0146] Define the objective function based on the matrix. :

[0147]

[0148] Where x is the actual response waveform to be recovered;

[0149] The objective function is solved analytically to obtain:

[0150]

[0151] in Reconstruct the true response waveform for candidate 3. for transpose, It is the identity matrix;

[0152] It should be noted that when solving the objective function, it is necessary to expand the objective function, calculate its gradient with respect to x, and set the gradient to zero to obtain the normal equation:

[0153]

[0154] because ,matrix It is reversible, therefore we obtain an analytical solution:

[0155]

[0156] Where x is the solution, which is the reconstructed true response waveform of candidate 3. ;

[0157] S2.4 Calculate the scoring items based on the reconstruction results of the multi-candidate algorithm, including the reconvolution residual ratio. Baseline fluctuation ratio and peak concentration penalty :

[0158] Among them, the reconvolution residual ratio for:

[0159]

[0160] in Reconstruct the true response waveform for the m-th candidate;

[0161] Baseline fluctuation ratio for:

[0162]

[0163] in for The standard deviation of the front and rear steady-state regions, This is the event window for the c-th sensor channel;

[0164] Front and rear static steady zones Including the calm and stable zone before the arrival of the main peak and the calm zone after the wake wave ends This refers to the silent zone outside the effective influence range of the main peak, which theoretically should mainly consist of background noise, weak reflection residue, and instrument noise. The location of the main peak is determined by constructing an absolute value envelope and calculating a baseline noise reference. The larger of 25% of the main peak's peak value and four times the noise level is taken as the main peak's main body threshold. A search is performed to the left of the main peak location to find a position where the absolute value envelope is less than or equal to the main peak's main body threshold and subsequent consecutive points do not exceed the threshold, which is then used as the left boundary. Similarly, a search is performed to the right from the main peak to find the right boundary. A protection interval is set, taking the area before the main peak's left boundary as the stable zone. The right endpoint is used as the starting point of the recording window to determine the range of the front stable region, and similarly, the rear stable region is taken at the right boundary of the main peak. If the right endpoint of the front static stable region exceeds the right endpoint, then the front static stable region is set to an empty set, and the rear static stable region is set in the same way.

[0165] Peak concentration penalty for:

[0166]

[0167] The comprehensive score for the reconstruction results of each candidate algorithm is obtained by weighted summation of the scoring items. The candidate algorithm with the highest comprehensive score is selected as the optimal reconstruction result. The optimal reconstruction results include the HFCT channel, UHF channel, and ultrasound channel.

[0168] S3. Determine the equivalent radial source position of a single event based on the optimal reconstruction result and calculate the propagation distance for geometric correction. Construct a propagation transfer function to compensate for the geometric correction and obtain the channel-normalized waveform.

[0169] S3.1 Determine the equivalent radial source location for a single event based on the optimal ultrasound reconstruction results. :

[0170] For the i-th event, from the four optimal ultrasound reconstruction results Extract the arrival time of the first arrival exceeding the ultrasonic trigger threshold. And take the minimum value from the four arrival times. ;

[0171] according to Calculate the equivalent radial source location :

[0172]

[0173] in For the ultrasonic channel at the acoustic excitation angular frequency Phase constant at that point, For the sound school's standard time delay, This is a truncation function;

[0174] Calculate the propagation distance of the i-th event to the UHF and ultrasonic sensor based on the equivalent radial source location:

[0175]

[0176] in For propagation distance, only UHF and ultrasonic sensors are included. Local high-risk sectors The projection angle of the outer shell is defined as 0. The circumference radius of the sensor is calculated using polar angle conversion;

[0177] For HFCT sensors, a fixed value is taken. ;

[0178] S3.2 Define the position excitation correction factor and geometric diffusion compensation factor for the i-th event:

[0179]

[0180] in This is the position excitation correction factor. It is the geometric diffusion compensation factor;

[0181] Geometric corrections were performed on the UHF and ultrasonic sensor channels:

[0182]

[0183]

[0184] in This represents the optimal reconstruction results for the UHF and ultrasonic sensor channels after geometric correction. The optimal reconstruction result for the HFCT channel. To obtain the optimal reconstruction result for the HFCT channels, it is necessary to start from... Extract the optimal reconstruction result corresponding to the HFCT channel;

[0185] S3.3, By repeatedly injecting into the piezoelectric calibration actuator The same stimulus was used again, and the calibration response was collected. And on The calibration frequency domain is obtained by averaging the responses synchronously and then performing a Fourier transform. The ultrasonic sensor channel with the shortest propagation distance was selected as the reference channel, and the reference calibration frequency domain was obtained. Define the relative propagation transfer function for any non-reference channel. :

[0186]

[0187] in To calibrate the frequency domain, For reference calibration frequency domain;

[0188] By propagating the transfer function Obtain the parameters of the j-th ultrasound channel, including the attenuation constant. and phase constant :

[0189]

[0190]

[0191] in Let be the propagation distance of the j-th ultrasonic sensor channel. For the reference channel propagation distance, This indicates taking the phase angle;

[0192] The parameters of all ultrasound channels are weighted and synthesized into general parameters. Specifically, the calibration signal-to-noise ratio of the j-th channel is calculated, the calibration signal-to-noise ratio is normalized and used as the synthesis weight, and the parameters of all ultrasound channels are weighted and synthesized into general attenuation constant and general phase constant.

[0193] S3.4. Based on general parameters, form the propagation transfer function of the UHF and ultrasonic sensor channels. The geometric correction results for the UHF and ultrasonic sensor channels are compensated.

[0194]

[0195] in The normalized spectrum after propagation compensation for UHF and ultrasonic sensor channels. This is the spectrum of the optimal reconstruction result after geometric correction. For propagation of the transfer function conjugate, To compensate for the regularization coefficient during propagation;

[0196] For HFCT channel fixation , For the normalized spectrum of the HFCT channel, The spectrum of the optimal reconstruction result of the HFCT channel, through the optimal reconstruction result of the HFCT channel. Obtain;

[0197] The compensated channel normalized waveform is obtained by using inverse Fourier transform. ;

[0198] Calculate the area response of each channel for the i-th event. :

[0199]

[0200] in include , as well as , With only one channel, the calculation is straightforward. There are two channels, and the area response values ​​from the two channels need to be averaged to obtain the result. There are four channels, and the area response of the four channels needs to be averaged to obtain the result.

[0201] The mean comprehensive scores for HFCT, UHF, and ultrasound channels are calculated based on the maximum comprehensive score for each channel. Further, an exponential function is used to calculate the reliability weights for HFCT, UHF, and ultrasound channels respectively. These reliability weights are then normalized. Based on these normalized reliability weights, the area-type response quantities are weighted and fused to obtain the fused response quantity for the i-th event. ;

[0202] It should be noted that after obtaining the maximum comprehensive score for each channel, the average comprehensive scores of the HFCT, UHF, and ultrasound channels are calculated separately. The average comprehensive score of each channel is input into the exponential function exp to obtain the reliability weights of the HFCT, UHF, and ultrasound channels. The reliability weight of each channel is normalized by dividing it by the sum of the reliability weights of the three channels. Based on the normalized reliability weight of each channel, the corresponding area-type response of each channel is multiplied, and finally the results are added together to obtain the fusion response.

[0203] S3.5. Map the fused response quantity to an uncalibrated equivalent response curve distributed along a local radial direction. :

[0204]

[0205] in The radial spread factor is... , The value range is 0.08~0.15, with a preferred value of 0.1. If the value is too small, the equivalent response will be excessively concentrated into a peak, causing the stress reconstruction to be overly sensitive to single events. If the size is too large, it will expand the local defect into an excessively wide distribution, weakening the sensitivity to localization; therefore, it is limited. Range of values;

[0206] During the device calibration phase, a standard response event obtained using a known excitation method, or the standard sample event selected from actual operating samples that has the highest signal-to-noise ratio and the best cross-channel consistency, will be used as the reference event. A reference equivalent response curve will be constructed based on the reference standard data of the reference event. And calculate the uniform calibration coefficient. ;

[0207]

[0208] in and For the integration interval near the reference boundary, , , For reference surface charge response, The relative permittivity of the main insulator. It is the vacuum permittivity;

[0209] Specifically, based on the reference standard data of the standard sample events in the reference events, including the reference fused response and the reference equivalent radial source location, the reference equivalent response curve is obtained by mapping the reference fused response and the reference equivalent radial source location using the formula of the uncalibrated equivalent response curve. ;

[0210] The equivalent response charge of the i-th event after calibration is obtained based on the unified calibration coefficient. and equivalent response density :

[0211]

[0212] For the i-th event, establish the radial first-order differential equation:

[0213]

[0214] in Let be the equivalent insulation stress response distribution for the i-th event;

[0215] The equivalent insulation stress response distribution is obtained by solving the first-order radial differential equation. :

[0216]

[0217] in For integration variables, It is the integration constant;

[0218] To ensure that the equivalent stress response is consistent with the local equivalent peak voltage, boundary constraints are applied:

[0219]

[0220] The integration constants can be determined through boundary constraints. ;

[0221] Specifically, the solution process for the radial first-order differential equation includes:

[0222] For a radial first-order differential equation, which is a first-order linear ordinary differential equation, take the integration factor. :

[0223]

[0224] Multiplying both sides of the radial first-order differential equation by the integrating factor yields:

[0225]

[0226] Further organized as follows:

[0227]

[0228] From the above formula Integrate up to r:

[0229]

[0230] make ,available:

[0231]

[0232] To obtain a specific solution that satisfies the local equivalent peak voltage condition, boundary constraints are given, and the equivalent insulation stress response distribution is substituted into the boundary constraints:

[0233]

[0234] The results were:

[0235]

[0236] because:

[0237]

[0238] Therefore, a unique integral constant can be obtained. :

[0239]

[0240] The only integral constant Substituting the equivalent insulation stress response distribution again yields a unique result. ;

[0241] exist to Within the range, the largest equivalent insulation stress response distribution is selected as the peak insulation stress for the i-th event. The enhancement factor is calculated based on the peak insulation stress. :

[0242]

[0243] in The maximum calibration field strength is obtained through electrical calibration excitation;

[0244] Specifically, the electrical calibration pulse is coupled to the local insulation channel through the calibration coupling electrode. The electrical calibration coupling coefficient is defined to map the electrical calibration excitation voltage to the equivalent internal excitation. The equivalent field strength is calculated through the equivalent internal excitation, and the maximum calibration field strength is obtained from the equivalent field strength.

[0245] S4. Normalize the channel standardized waveforms to construct the main cluster waveform signature of the events, calculate the cluster distance to perform clustering, map the events to the phase window index and spatial index, and combine the clustering results to construct a weighted event set;

[0246] S4.1. Normalize the energy of the standardized waveforms of the two UHF sensor channels for the i-th event, and exponentially calculate the fusion weight of the two UHF sensor channels by using the maximum comprehensive score of each UHF sensor channel. Construct the main cluster waveform signature for the i-th event based on the fusion weight. :

[0247]

[0248] in and The normalized waveforms of the energy from the two UHF sensor channels are shown. and The fusion weights for the two UHF sensor channels;

[0249] Define window For windows For any two events i and j within the range, calculate the elements of the globally normalized cross-correlation matrix. :

[0250]

[0251] in For the time lag, Sign the main cluster waveform for event j;

[0252] The elements of the globally normalized cross-correlation matrix are used to form a similarity matrix. And convert it into the main distance matrix. :

[0253]

[0254] Based on the principal distance matrix Calculate the cluster distance between events i and j :

[0255]

[0256] in Elements of the principal distance matrix For the amplification factor of event j, To prevent division by zero decimals, , , For distance weights, satisfying ;

[0257] Cluster distances are used to form a distance matrix, and average connectivity hierarchical clustering is used for clustering. The clustering stopping condition is set to the point where the average distance between any two clusters is greater than a threshold. Stop merging when the clustering is complete and output the clustering results.

[0258] S4.2, Calculation Window The power frequency phase of the i-th event :

[0259]

[0260] in This is the main triggering time of the i-th event. The starting point of the zero-phase power frequency within the window. The power frequency cycle;

[0261] Based on the power frequency phase, 12 equal-width phase windows are used to map the i-th event to the phase window index. :

[0262]

[0263] in The number of equal-width phase windows, The width of each equal-width phase window;

[0264] Local equivalent insulation channel The space is fixedly divided into four radial regions of equal width, and the spatial index of the i-th event is defined. :

[0265]

[0266] in The number of radially spaced regions of equal width. The width of each equal-width radial space region;

[0267] Based on phase window index Spatial Index And construct a weighted event set from the clustering results. :

[0268]

[0269] in This is the cluster index for the i-th event.

[0270] S5. Define the total intensity of the phase cluster and the spatial concentration factor within the cluster based on the weighted event set, and define the joint fingerprint unit and the total joint coupling strength;

[0271] S5.1 Calculate the dual UHF average score of the i-th event and define the physical consistency factor of the i-th event relative to the cluster mean. :

[0272]

[0273] in For the cluster Average radial position, For the cluster The average stress enhancement factor, To prevent extremely small positive numbers with a denominator of zero, take , and The coefficient for uniformity suppression is preferably taken as follows: The value is 1.5 because radial positional deviation directly reflects whether it may originate from the same local defect area; events with inconsistent positions should be significantly suppressed. We set it to 1 because the stress enhancement factor will drift to some extent with changes in working conditions, and it cannot suppress excessive stress. If the value is too low, then outlier events that cross spatial regions will be incorrectly preserved. If the value is too high, the same defect may be mistakenly identified as an outlier event during the evolution of strength and weakness. exp is an exponential function. For the equivalent radial source location, For the equivalent inner radius, For the equivalent outer radius, Let be the amplification factor for event i;

[0274] Specifically, By belonging to the cluster The enhancement factor of all events is calculated by averaging the values.

[0275] Event credibility weights are calculated based on physical consistency factors and average scores from both UHF systems. And define the equivalent discharge amplitude. :

[0276]

[0277] in The average score for dual UHF signals is calculated by averaging the scores of the optimal reconstruction results from the two UHF sensor channels. The equivalent response charge, The waveform suppression coefficient is preferably set to 2, so that events with poor scores are significantly suppressed but not completely eliminated.

[0278] For weighted event set Each group Define weighted discharge intensity :

[0279]

[0280] in This represents the number of power frequency cycles within the current statistical window.

[0281] S5.2. The total intensity of the phase cluster is obtained by summing the weighted discharge intensities of the radial spatial region for the same phase window and cluster. And calculate the spatial concentration factor within the cluster. :

[0282]

[0283] Simultaneous calculation of weighted average peak stress for the same phase window and cluster :

[0284]

[0285] Calculate the local equivalent temperature based on the temperature sensor. Constructing phase-cluster stress enhancement factor :

[0286]

[0287]

[0288] in The temperature fusion coefficient, The preferred reference temperature is 25℃. The preferred reference field strength is the maximum calibration field strength. and These are the temperature sensitivity coefficient and the electric field sensitivity coefficient;

[0289] S5.3 Define a joint fingerprint unit for each phase window p and cluster e. :

[0290]

[0291] in The spatial concentration enhancement coefficient is preferably 0.40, which is the optimal balance between the ability to express spatial distribution differences and robustness.

[0292] The total joint coupling strength is obtained by summing the joint fingerprint units. .

[0293] S6. Establish a multi-dimensional feature vector based on the joint fingerprint unit and the total joint coupling strength, and standardize it. Identify the defect category of the voltage transformer based on the standardized multi-dimensional feature vector.

[0294] S6.1, Define multidimensional feature vectors :

[0295]

[0296] in For the total strength of joint coupling , Mean temperature-field enhancement factor , Average spatial concentration factor , Peak stress of the window ;

[0297] Specifically, mean temperature-field enhancement factor for:

[0298]

[0299] Average spatial concentration factor for:

[0300]

[0301] Peak stress of window The maximum peak insulation stress of the events in the window;

[0302] Median-interquartile range robust normalization is used for multidimensional feature vectors The process yields a standardized multidimensional feature vector. , respectively corresponding to the multidimensional feature vectors The standardization results;

[0303] Specifically, the robustly standardized median-interquartile range is represented as:

[0304]

[0305] in For the i-th standardized feature, including , For the i-th feature, including , The baseline median of the corresponding feature in the reference sample library, The baseline interquartile range of the corresponding feature in the reference sample library, To prevent numerical instability caused by excessively small interquartile ranges, the following protection term is adopted: The reference sample library can be obtained through actual running sample data or through offline labeled samples for calibration;

[0306] S6.2 Define six types of defect systems, including normal state, internal discharge type defects, surface discharge type defects, moisture / medium deterioration type defects, internal discharge and moisture composite defects and surface discharge and moisture composite defects.

[0307] For each defect category, a linear discriminant score is used. :

[0308]

[0309] in For the i-th standardized feature, Let be the weight of the i-th standardized feature under the n-th defect category. The bias is for the nth defect category;

[0310] Specifically, weight and bias Training is required. First, training data is obtained from historical data or a database. This training data is then labeled. Predicted scores are calculated using the training data against the linear discriminant scores. The predicted scores are then compared to the labeled scores in the training data to calculate the cross-entropy loss. Based on this cross-entropy loss, the trainer is optimized to adjust the weights. and bias Perform iterative optimization, recalculating the loss after each iteration. When the loss converges after each iteration, output the optimal weights. and bias ;

[0311] Linear discriminant score The probability of each defect category is obtained by using softmax normalization, and the defect category with the highest probability is selected as the final defect.

[0312] This embodiment also provides a voltage transformer discharge detection system, including:

[0313] The data acquisition module is used to identify local high-risk insulation sectors of the voltage transformer and establish a radial reference field. It collects multi-source data through sensors and performs calibration excitation modeling.

[0314] The event extraction module is used to build a discrete event observation model based on multi-source data and convert it into a frequency domain representation. It also builds a threshold to extract events based on sensors and outputs event windows. For each event window, it uses a multi-candidate algorithm to reconstruct the results and calculates a comprehensive score to select the optimal reconstruction result.

[0315] The geometry correction module is used to determine the equivalent radial source position of a single event based on the optimal reconstruction result and calculate the propagation distance for geometry correction. It also constructs a propagation transfer function to compensate for the geometry correction and obtain the channel-normalized waveform.

[0316] The clustering module is used to normalize the channel-standardized waveforms, construct the master cluster waveform signature of events, calculate the cluster distance, perform clustering, map events to phase window indices and spatial indices, and construct a weighted event set by combining the clustering results.

[0317] The defect analysis module is used to define the total intensity of phase clusters and the spatial concentration factor within the clusters based on the weighted event set, and to define the joint fingerprint unit and the total joint coupling intensity. It establishes a multi-dimensional feature vector based on the joint fingerprint unit and the total joint coupling intensity for standardization, and identifies the defect category of the voltage transformer based on the standardized multi-dimensional feature vector.

[0318] This embodiment also provides a computer device applicable to the voltage transformer discharge detection method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the voltage transformer discharge detection method proposed in the above embodiment.

[0319] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0320] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the voltage transformer discharge detection method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0321] In summary, this invention, by identifying local high-risk insulation sectors of voltage transformers and establishing a radial reference field, combined with multi-source sensing acquisition and calibration excitation modeling, transforms the discharge observation range from coarse monitoring to directional constraint of local insulation-sensitive areas, thus improving the targeting of detection from the source. By establishing a discrete event observation model, extracting event windows, and optimizing multi-candidate reconstruction, robust recovery of discharge events under complex noise backgrounds is achieved, providing reliable input for subsequent analysis. Through single-event equivalent radial source localization, propagation distance geometric correction, and propagation compensation, more comparable channel-standardized waveforms are obtained, which helps to reduce the influence of path differences and sensing differences. Furthermore, through clustering, phase-space mapping, and weighted event set construction, the characterization of discharge activity is improved from single-point features to cluster behavior features. Finally, by combining joint fingerprint units and joint coupling total strength, multi-dimensional defect identification is completed, thereby improving the accuracy, stability, and engineering applicability of voltage transformer defect classification.

[0322] It should be noted that 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting discharge in a voltage transformer, characterized in that: include, Identify local high-risk insulation sectors of voltage transformers and establish radial reference fields; collect multi-source data through sensors and perform calibration excitation modeling. A discrete event observation model is established based on multi-source data and converted into a frequency domain representation. A threshold is established based on sensors to extract events and output event windows. For each event window, a multi-candidate algorithm is used to reconstruct the results and a comprehensive score is calculated to select the optimal reconstruction result. Based on the optimal reconstruction results, the equivalent radial source position of a single event is determined and the propagation distance is calculated for geometric correction. The propagation transfer function is constructed to compensate for the geometric correction and obtain the channel-normalized waveform. The channel-standardized waveforms are normalized to construct the main cluster waveform signature of the events. The clustering distance is calculated to perform clustering. The events are mapped to the phase window index and spatial index. The weighted event set is constructed by combining the clustering results. The total intensity of the phase cluster and the spatial concentration factor within the cluster are defined based on the weighted event set, and the joint fingerprint unit and the total joint coupling strength are also defined. A multi-dimensional feature vector is established based on the joint fingerprint unit and the total joint coupling strength and then standardized. Based on the standardized multi-dimensional feature vector, the defect category of the voltage transformer is identified.

2. The voltage transformer discharge detection method as described in claim 1, characterized in that: The process involves identifying local high-risk insulation sectors of the voltage transformer and establishing a radial reference field, acquiring multi-source data through sensors and performing calibration excitation modeling to obtain voltage transformer structural data and select local high-risk insulation sectors. In the insulated mountain area, the average outer radius of the outer high-voltage winding is defined as the equivalent inner radius. And define the equivalent outer radius ; Define local equivalent peak voltage And establish a radial reference field for a locally axisymmetric equivalent insulation channel. ; Multi-source sensors are deployed on the voltage transformer to collect multi-source data, including 1 HFCT sensor, 2 UHF sensors, 4 ultrasonic sensors, 2 temperature sensors, and 1 power frequency reference voltage sampling terminal. In local high-risk insulation sectors The calibration excitation model was modeled using an electrical calibration excitation actuator and a piezoelectric acoustic calibration actuator.

3. The voltage transformer discharge detection method as described in claim 2, characterized in that: The process of converting a discrete event observation model based on multi-source data into a frequency domain representation, extracting event windows based on sensor thresholds, reconstructing the results of each event window using a multi-candidate algorithm, calculating a comprehensive score, and selecting the optimal reconstruction result includes: A discrete event observation model is established based on the collected multi-source data, and the discrete event observation model is converted into a frequency domain representation. The number of sampling points is defined for the HFCT sensor channel. The front window is used to calculate the mean within the front window. and standard deviation HFCT trigger thresholds were established based on the mean and standard deviation. ; When HFCT sensor channel data When the time is greater than or equal to the trigger threshold, Defined as the main trigger time of the i-th candidate event, and based on time... Build Event Window ; The validity of candidate events is further determined based on the UHF and ultrasonic sensors, and an event window is output. ; The result is reconstructed using a multi-candidate algorithm for the i-th event window; The scoring criteria, including the reconvolution residual ratio, are calculated based on the reconstruction results of the multi-candidate algorithm. Baseline fluctuation ratio and peak concentration penalty ; The comprehensive score for the reconstruction results of each candidate algorithm is obtained by weighted summation of the scoring items. The candidate algorithm with the highest comprehensive score is selected as the optimal reconstruction result. .

4. The voltage transformer discharge detection method as described in claim 3, characterized in that: The process of determining the equivalent radial source position of a single event based on the optimal reconstruction result, calculating the propagation distance for geometric correction, and constructing a propagation transfer function to compensate for the geometric correction to obtain the channel-normalized waveform includes: The equivalent radial source location for a single event was determined based on the optimal ultrasound reconstruction results. ; The propagation distance of the i-th event to the UHF and ultrasonic sensor is calculated based on the equivalent radial source location; Define a position excitation correction factor and a geometric diffusion compensation factor for the i-th event; Geometric correction of UHF and ultrasonic sensor channels; By repeatedly injecting into the piezoelectric calibration actuator The same stimulus was used again, and the calibration response was collected. And on The calibration frequency domain is obtained by averaging the responses synchronously and then performing a Fourier transform. The ultrasonic sensor channel with the shortest propagation distance was selected as the reference channel, and the reference calibration frequency domain was obtained. Define the relative propagation transfer function for any non-reference channel. ; By propagating the transfer function Obtain the parameters of the j-th ultrasound channel; The parameters of all ultrasound channels are weighted and synthesized into a common parameter. Based on the common parameter, the propagation transfer function of the UHF and ultrasound sensor channels is formed. And compensate for the geometric correction results of the UHF and ultrasonic sensor channels; The compensated channel normalized waveform is obtained by using inverse Fourier transform. ; Calculate the area response of each channel for the i-th event. ; The mean comprehensive scores for HFCT, UHF, and ultrasound channels are calculated based on the maximum comprehensive score for each channel. Further, an exponential function is used to calculate the reliability weights for HFCT, UHF, and ultrasound channels respectively. These reliability weights are then normalized. Based on these normalized reliability weights, the area-type response quantities are weighted and fused to obtain the fused response quantity for the i-th event. ; The fused response is mapped to an uncalibrated equivalent response curve distributed along a local radial direction. ; Construct a reference equivalent response curve And calculate the uniform calibration coefficient. ; The equivalent response charge of the i-th event after calibration is obtained based on the unified calibration coefficient. and equivalent response density ; For the i-th event, a radial first-order differential equation is established, and the equivalent insulation stress response distribution is obtained by solving the radial first-order differential equation. ; exist to Within the range, the largest equivalent insulation stress response distribution is selected as the peak insulation stress for the i-th event. The enhancement factor is calculated based on the peak insulation stress. .

5. The voltage transformer discharge detection method as described in claim 4, characterized in that: The process involves normalizing the channel-standardized waveforms to construct the master cluster waveform signature for the event, calculating the cluster distance, and performing energy normalization on the standardized waveforms of the two UHF sensor channels for the i-th event. The maximum comprehensive score of each UHF sensor channel is then exponentially calculated to determine the fusion weight of the two UHF sensor channels. Based on this fusion weight, the master cluster waveform signature for the i-th event is constructed. ; Define window For windows For any two events i and j within the range, calculate the elements of the globally normalized cross-correlation matrix. ; The elements of the globally normalized cross-correlation matrix are used to form a similarity matrix. And convert it into the main distance matrix. ; Based on the principal distance matrix Calculate the cluster distance between events i and j ; Cluster distances are used to form a distance matrix, and average connectivity hierarchical clustering is used for clustering. The clustering stopping condition is set to the point where the average distance between any two clusters is greater than a threshold. Stop merging when the clustering is complete and output the clustering results.

6. The voltage transformer discharge detection method as described in claim 5, characterized in that: The process involves mapping events to phase window indices and spatial indices, and constructing a weighted event set based on clustering results, which is then used to calculate the window. The power frequency phase of the i-th event ; Based on the power frequency phase, 12 equal-width phase windows are used to map the i-th event to the phase window index. ; Local equivalent insulation channel The space is fixedly divided into four radial regions of equal width, and the spatial index of the i-th event is defined. ; Based on phase window index Spatial Index And construct a weighted event set from the clustering results. .

7. The voltage transformer discharge detection method as described in claim 6, characterized in that: The process involves defining the total intensity of the phase cluster and the spatial concentration factor within the cluster based on the weighted event set, defining the joint fingerprint unit and the joint coupling total strength to calculate the dual UHF average score of the i-th event, and defining the physical consistency factor of the i-th event relative to the mean of its cluster. ; Event credibility weights are calculated based on physical consistency factors and average scores from both UHF systems. And define the equivalent discharge amplitude. ; For weighted event set Each group Define weighted discharge intensity ; The total intensity of the phase cluster is obtained by summing the weighted discharge intensities in the radial spatial region for the same phase window and cluster. And calculate the spatial concentration factor within the cluster. ; Simultaneous calculation of weighted average peak stress for the same phase window and cluster ; Calculate the local equivalent temperature based on the temperature sensor. Constructing phase-cluster stress enhancement factor ; Define a joint fingerprint unit for each phase window p and cluster e. ; The total joint coupling strength is obtained by summing the joint fingerprint units. .

8. The voltage transformer discharge detection method as described in claim 7, characterized in that: The standardization of the multi-dimensional feature vector established based on the joint fingerprint unit and the total joint coupling strength, and the identification of voltage transformer defect categories based on the standardized multi-dimensional feature vector, include: Define multidimensional feature vectors Median-interquartile range robust normalization is used for multidimensional feature vectors. The process yields a standardized multidimensional feature vector. ; A six-category defect system is defined, and for each defect category, a linear discriminant score is used. ; Linear discriminant score The probability of each defect category is obtained by using softmax normalization, and the defect category with the highest probability is selected as the final defect.

9. A voltage transformer discharge detection system, based on the voltage transformer discharge detection method according to any one of claims 1 to 8, characterized in that: include, The data acquisition module is used to identify local high-risk insulation sectors of the voltage transformer and establish a radial reference field. It collects multi-source data through sensors and performs calibration excitation modeling. The event extraction module is used to build a discrete event observation model based on multi-source data and convert it into a frequency domain representation. It also builds a threshold to extract events based on sensors and outputs event windows. For each event window, it uses a multi-candidate algorithm to reconstruct the results and calculates a comprehensive score to select the optimal reconstruction result. The geometry correction module is used to determine the equivalent radial source position of a single event based on the optimal reconstruction result and calculate the propagation distance for geometry correction. It also constructs a propagation transfer function to compensate for the geometry correction and obtain the channel-normalized waveform. The clustering module is used to normalize the channel-standardized waveforms, construct the master cluster waveform signature of events, calculate the cluster distance, perform clustering, map events to phase window indices and spatial indices, and construct a weighted event set by combining the clustering results. The defect analysis module is used to define the total intensity of phase clusters and the spatial concentration factor within the clusters based on the weighted event set, and to define the joint fingerprint unit and the total joint coupling intensity. It establishes a multi-dimensional feature vector based on the joint fingerprint unit and the total joint coupling intensity for standardization, and identifies the defect category of the voltage transformer based on the standardized multi-dimensional feature vector.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the voltage transformer discharge detection method according to any one of claims 1 to 8.

Images