An intelligent detection platform and method for transformer PD sources with ultra-wide frequency band

Abstract

The present application discloses an intelligent detection platform and method for transformer PD sources with ultra-wide frequency bands, relating to the field of electrical variable measurement, including: a configuration module for configuring a sensor group, an initial solution construction module for obtaining wavelet decomposition coefficients at each scale after performing wavelet decomposition on the pulse signal set, a correlation analysis module for calculating the particle similarity of m initial particles, an update and iteration module for performing correlation selection on the particle correlations and generating a correlation optimization based on the current iteration number and iteration status, a denoising processing module for dynamically adjusting the inertia weight, a feature extraction module for establishing a feature extraction result for the denoised signal set, and an anomaly analysis module for performing transformer anomaly analysis based on the feature extraction result and generating an anomaly report. It solves the technical problem of low accuracy in distinguishing ultra-wide frequency band PD signals from noise sources, and achieves the technical effect of extracting PD pulse characteristic parameters and realizing intelligent clustering and separation.

Description

Technical Field

[0001] The present application relates to the field of electrical variable measurement, and in particular, to an intelligent detection platform and method for a transformer PD source with an ultra-wide frequency band. Background Art

[0002] With the rapid construction and development of the smart grid in China, the reliability and security of power grid operation have been gradually improved. As key power transmission and transformation equipment in AC and DC systems, the operating status of power transformers and converter transformers directly affects the stable operation of the entire power grid. At present, oil-paper composite insulation is a common insulation method for large transformers, which is mainly composed of insulating oil, insulating paperboard, and other solid insulating materials. Although oil-paper insulation is required to have sufficient electrical strength and mechanical properties in design, it is inevitable that accidental factors in the production, installation, and operation processes cause defects in the insulation system, leading to equipment failures. Partial discharge (PD) is the main cause of transformer oil-paper insulation defect failures and can be used as a means to evaluate the degree of insulation deterioration. Therefore, carrying out PD detection (monitoring) of power transformers and converter transformers under AC and DC voltages and obtaining discharge information will help to master the insulation status of in-service transformers and ensure their safe and stable operation. Summary of the Invention

[0003] By providing an intelligent detection platform and method for a transformer PD source with an ultra-wide frequency band, the present application solves the technical problem of low accuracy in distinguishing ultra-wide frequency band PD signals from noise sources, and achieves the technical effect of quickly extracting PD pulse characteristic parameters and realizing intelligent clustering and separation.

[0004] The present application provides an intelligent detection platform for ultra-wideband transformer PD sources, and the platform is applied to an intelligent detection method for ultra-wideband transformer PD sources, including: a configuration module for configuring a sensor group, the sensor group including HFCT, a bushing end screen sensor, and a detection impedance, monitoring the partial discharge pulse signals of the transformer by using the sensor group, and establishing a pulse signal set; an initial solution construction module for performing wavelet decomposition on the pulse signal set, obtaining wavelet decomposition coefficients at each scale, and generating m initial particles in the solution space by using the wavelet decomposition coefficients; a correlation analysis module for calculating the particle similarity of the m initial particles, establishing a particle similarity correlation, performing particle combination by using the particle similarity correlation, and establishing a particle correlation; an update and iteration module for performing correlation selection on the particle correlation, generating a correlation optimization by using the current iteration number and iteration state, and updating and iterating the m initial particles by using the correlation optimization, the individual extreme value of the particle, and the population extreme value of the particle; a denoising processing module for dynamically adjusting the inertia weight, performing iterative optimization, establishing an optimal wavelet threshold according to the iterative optimization result, performing denoising processing on the pulse signal set by using the optimal wavelet threshold, and establishing a denoised signal set; a feature extraction module for extracting first-order shape parameter features, kurtosis features, skewness features, and frequency domain features from the denoised signal set, and establishing a feature extraction result; an anomaly analysis module for performing anomaly analysis on the transformer according to the feature extraction result and generating an anomaly report.

[0005] In a possible implementation manner, the correlation analysis module performs the following processing: configuring a similarity calculation formula as follows: ; where represents the particle similarity correlation between particle and particle , and respectively represent fitness functions for measuring the individual advantages and disadvantages of particle and particle , is the particle fitness deviation coefficient, and is the particle self-deviation coefficient; configuring a combination isolation strategy, after sorting the correlation order of the particle similarity correlation, performing correlation screening from the maximum correlation by using the combination isolation strategy to complete particle combination and establish a particle correlation.

[0006] In a possible implementation, the update iteration module performs the following processing: After performing correlation normalization on the particle correlations, perform downward correlation ratio clustering starting from the highest correlation value to establish a downward correlation ratio clustering result; when the downward correlation ratio clustering result meets a preset threshold, the clustering ends, and the particles corresponding to the particle correlations in the clustering are used as correlation selection particles; obtain the current iteration count, generate a first optimization weight based on the iteration count; obtain the current iteration state, generate a second optimization weight based on the iteration state; after comprehensively calculating the first optimization weight and the second optimization weight, generate an optimization reward using the low fitness particles in the correlation selection particles and the comprehensive calculation result, update the high fitness particles in the correlation selection particles using the optimization reward, and stop the update of the low fitness particles, and complete the update iteration according to the update result.

[0007] In a possible implementation, the update iteration module performs the following processing: Remove the low fitness particles that have stopped updating from the m initial particles to establish updated initial particles; perform individual updates on the non-correlation selection particles in the updated initial particles based on the particle individual extreme value and the particle swarm extreme value to generate a first update result; perform individual updates on the correlation selection particles in the updated initial particles based on the optimization reward, the particle individual extreme value, and the particle swarm extreme value to generate a second update result; after supplementing new particles to the first update result and the second update result, complete the update iteration.

[0008] In a possible implementation, the feature extraction module performs the following processing: Calculate the kurtosis feature through the formula as follows: ; where represents kurtosis, represents the th value of the random variable, represents the mean of the random variable, is the standard deviation of the random variable, is the total number of random variables, represents the value of the probability density function of the random variable at ; Calculate the skewness feature through the formula as follows: ; where represents skewness; The kurtosis feature and the skewness feature are used as the feature extraction results.

[0009] In a possible implementation, the anomaly analysis module performs the following processing: collecting device status parameters of the transformer to establish a device status data set, where the device status data set includes load data, voltage level data, and operating time data; collecting environmental parameters of the transformer to establish an environmental data set, where the environmental data set includes temperature data and humidity data; using the device status data set, the environmental data set, and the feature extraction result to identify transformer anomalies and establish an anomaly report.

[0010] In a possible implementation, the following processing is performed: a monitoring module for establishing a monitoring window according to the anomaly report; a feedback monitoring module for performing feedback update on the feature extraction module based on the monitoring window, and performing transformer monitoring and management according to the feature extraction module after feedback update.

[0011] The present application also provides an ultra-wideband intelligent detection method for transformer PD sources, including: configuring a sensor group, where the sensor group includes HFCT, a bushing end screen sensor, and a detection impedance, using the sensor group to monitor partial discharge pulse signals of the transformer to establish a pulse signal set; after performing wavelet decomposition on the pulse signal set, obtaining wavelet decomposition coefficients at each scale, and generating m initial particles in the solution space using the wavelet decomposition coefficients; calculating particle similarity of the m initial particles to establish particle similarity associations, and using the particle similarity associations to perform particle combination to establish particle associations; after performing association selection on the particle associations, generating association optimization through the current iteration number and iteration status, and using the association optimization, particle individual extreme values, and particle swarm extreme values to update and iterate the m initial particles; dynamically adjusting the inertia weight, performing iterative optimization, establishing an optimal wavelet threshold according to the iterative optimization result, using the optimal wavelet threshold to perform denoising processing on the pulse signal set, and establishing a denoised signal set; extracting first-order shape parameter features, kurtosis features, skewness features, and frequency domain features from the denoised signal set to establish a feature extraction result; performing transformer anomaly analysis according to the feature extraction result to generate an anomaly report.

[0012] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0013] An ultra-wideband intelligent detection platform and method for transformer PD sources provided by the present application relate to the technical field of electrical variable measurement, solve the technical problem of low accuracy in distinguishing ultra-wideband PD signals from noise sources, and achieve the technical effect of quickly extracting PD pulse characteristic parameters and realizing intelligent clustering and separation. Description of the Drawings

[0014] To more clearly illustrate the technical solutions of the embodiments of the present application, the accompanying drawings of the embodiments of the present application will be briefly introduced below. Flowcharts are used in the present application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the operations above or below do not necessarily need to be performed precisely in sequence. On the contrary, as needed, various steps can be performed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or more steps can be removed from these processes.

[0015] Figure 1 It is a schematic structural diagram of an intelligent detection platform for ultra-wideband transformer PD sources provided by an embodiment of the present application.

[0016] Figure 2 It is a schematic diagram of time-domain characteristic parameters of a single discharge waveform of an intelligent detection platform for ultra-wideband transformer PD sources provided by an embodiment of the present application.

[0017] Figure 3 It is a schematic flowchart of an intelligent detection method for ultra-wideband transformer PD sources provided by an embodiment of the present application. Detailed implementation manners

[0018] The above description is only an overview of the technical solutions of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the following specifically gives the detailed implementation manners of the present application.

[0019] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations of the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0020] In the following description, "some embodiments" are involved, which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. The terms "first / second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application.

[0021] An intelligent detection platform for ultra-wideband transformer PD sources according to an embodiment of this application is as follows Figure 1 shown, and the method includes:

[0022] An intelligent detection platform for ultra-wideband transformer PD sources according to an embodiment of this application is used to solve the technical problem of low accuracy in distinguishing ultra-wideband PD signals from noise sources, and achieves the technical effect of quickly extracting PD pulse characteristic parameters and realizing intelligent clustering and separation. An intelligent detection platform for ultra-wideband transformer PD sources includes: a configuration module 10, an initial solution construction module 20, a correlation analysis module 30, an update and iteration module 40, a denoising processing module 50, a feature extraction module 60, and an anomaly analysis module 70.

[0023] The configuration module 10 is used to configure a sensor group, and the sensor group includes HFCT, bushing end screen sensors, and detection impedance. The sensor group is used to monitor the partial discharge pulse signals of the transformer and establish a pulse signal set. Specifically, in the partial discharge monitoring of the transformer, signal acquisition and analysis are carried out by configuring the sensor group. The configured sensor group can include HFCT, bushing end screen sensors, detection impedance, etc. The HFCT is used to capture high-frequency current signals outside the transformer, especially high-frequency current signals generated by partial discharge. The bushing end screen sensor is used to monitor partial discharge signals, especially the electric field and partial current outside the transformer, and can more accurately detect the electric field changes generated by partial discharge. The detection impedance can judge the partial discharge situation by detecting the change of the insulation impedance of the transformer. Partial discharge is usually accompanied by the aging and deterioration of insulating materials, and the impedance of the transformer will change with the occurrence of discharge.

[0024] Furthermore, the HFCT can be installed outside the transformer, especially at the input or output ends of the current conductors, and can monitor the changes in external current. The bushing end screen sensor is installed at the end of the transformer bushing, facilitating the detection of signals caused by local current or electric field of the transformer. The detection impedance sensor is usually installed in the winding or grounding part of the transformer to monitor the state of the insulating material.

[0025] At the same time, using the sensor group to monitor the partial discharge pulse signals of the transformer means starting the sensor group for signal monitoring. Each sensor will collect signals from the target area at a set time interval and transmit the collected signals to the data acquisition system, which can include real-time collection of high-frequency current signals through the HFCT sensor and sending the data to the acquisition system for further analysis and processing, continuously monitoring the local electric field changes through the bushing end screen sensor, recording the change conditions and transmitting them to the data processing unit, and then continuously monitoring the insulation state of the transformer through the monitoring impedance sensor, recording the impedance change data, and transmitting the signals to the data acquisition system through the signal, and ensuring that the data acquisition processes of all sensors are synchronized to ensure the accurate capture of partial discharge signals and avoid the influence of timing errors on data quality.

[0026] Finally, based on the processed signal data, signal processing algorithms such as fast Fourier transform (FFT) and time-frequency analysis can be used to extract the frequency, time, and spatial characteristics of the partial discharge signals. By classifying the partial discharge signals with different characteristics, different types of discharge signal sets are generated. For example, the discharge signals can be classified into categories such as mild discharge, normal discharge, and severe discharge according to the intensity and frequency of the discharge signals, and the integration of different sensor data forms a comprehensive partial discharge pulse signal set.

[0027] The initial solution construction module 20 is used to perform wavelet decomposition on the pulse signal set, obtain the wavelet decomposition coefficients at each scale, and generate m initial particles in the solution space; specifically, perform multi-layer wavelet decomposition on the pulse signal. Each layer of decomposition generates a low-frequency part (approximate coefficient) and a high-frequency part (detail coefficient). Usually, N layers of wavelet decomposition are performed to obtain signal components at different scales. Select an appropriate number of wavelet decomposition layers N, and the number of wavelet decomposition layers N is related to the frequency range of the signal and the desired resolution. Among them, a higher number of decomposition layers can extract more detailed signal characteristics.

[0028] Furthermore, wavelet decomposition generates coefficients of different scales, mainly including the low-frequency part (approximate coefficients) and the high-frequency part (detail coefficients). The low-frequency coefficients (approximate coefficients): reflect the long-term trend of the signal. The low-frequency part is the stable part of the signal and usually contains the global characteristics of the signal. The high-frequency coefficients (detail coefficients): reflect the local changes or instantaneous characteristics of the signal. The high-frequency part can capture the sharp changes, pulses and other short-term change characteristics in the signal. Each layer of wavelet decomposition generates a set of coefficients, and each set of coefficients is merged in turn to obtain the characteristics of the signal at different scales. After obtaining the wavelet decomposition coefficients at each scale, particles can be generated through the solution space. It means that the number and characteristic dimensions of the wavelet decomposition coefficients determine the dimension of the solution space. In the solution space, each particle corresponds to a potential solution, which consists of multiple dimensions, and each dimension represents a characteristic or coefficient value. For example, each wavelet decomposition coefficient can be regarded as a dimension of the particle, and at the same time, each particle is initialized by the wavelet decomposition coefficients. Specifically, for the coefficients obtained by each layer of decomposition, the position and velocity of the initial particle are generated. Further, the position of the initial particle can be directly associated with the wavelet decomposition coefficients of each layer. The dimension of the initial particle is equal to the number of decomposition layers multiplied by the number of wavelet coefficients in each layer, and the velocity of the particle is adjusted by the current coefficient value and the historical best solution to determine the initial velocity of the particle; according to the above-mentioned initialized particle velocity and particle position, m initial particles are searched in the solution space.

[0029] The association analysis module 30 is used to calculate the particle similarity of the m initial particles, establish the particle similarity association, and use the particle similarity association to perform particle combination to establish the particle association; below, the specific configuration of the association analysis module 30 will be described in detail. The association analysis module 30 may further include: a first configuration unit for configuring a similarity calculation formula as follows:

[0030] ;

[0031] Wherein, represents the particle similarity association between particle and particle , and respectively represent the fitness functions for measuring the individual advantages and disadvantages of particle and particle , is the particle fitness deviation coefficient, is the particle self-deviation coefficient; a second configuration unit for configuring a combination isolation strategy, after sorting the association order of the particle similarity association, performing association screening from the maximum association through the combination isolation strategy to complete the particle combination and establish the particle association.

[0032] Specifically, in the Particle Swarm Optimization (PSO) algorithm, the processes of calculating particle similarity and establishing particle similarity association can help improve the optimization efficiency and enhance the exploration ability of the particle swarm in the solution space. By calculating the similarity between particles, particles with similar characteristics can be found and grouped together for optimization, thus improving the search effect. The particle similarity is a measure of the similarity degree between particles. Usually, the position or velocity difference of particles is used as the basis for similarity calculation, and the similarity can be calculated by configuring a similarity calculation formula. The configuration of the similarity calculation formula is as follows:

[0033] ;

[0034] Among them, represents the particle similarity association between particle and particle . and respectively represent the fitness functions that measure the individual advantages and disadvantages of particle and particle . is the particle fitness deviation coefficient, is the particle self-deviation coefficient.

[0035] Traverse the similarities between m particles and store them in the similarity matrix. Further, group the particles according to the similarity and establish the structure of a particle swarm. By setting a similarity threshold, similar particles can be grouped into one group. The similarity threshold can be set as a fixed threshold or adjusted dynamically. For example, when the Euclidean distance between two particles is less than a certain set value, they are considered similar and can be grouped into the same group. At the same time, according to the similarity matrix and the set threshold T, all particles are divided into multiple sub-populations. The particles within each sub-population are similar, and it can be considered that they explore similar regions in the solution space. Through the similarity matrix, particles with higher similarity are grouped into one group to form multiple particle sub-populations. Each sub-population represents an exploration direction or a solution space region.

[0036] Further configure the combination isolation strategy. The combination isolation strategy means selecting multiple particles from the particle group with higher similarity for combination to maintain the diversity and effectiveness of the particle group. And the paired particles cannot be paired with other particles anymore to achieve the isolation purpose. After sequential screening from the maximum association, pairwise combinations of particles can be constructed, and the particles that cannot be combined are eliminated. After the particle combination is completed, update the solution space position of the particle group, calculate the new fitness value, and synchronize the updated particle group and the similarity association information to the next iteration of the particle swarm optimization process to establish particle association, thereby improving the optimization efficiency and the quality of the final solution.

[0037] The update and iteration module 40 is used to perform correlation selection on the particle correlation, generate correlation optimization through the current iteration number and iteration state, and update and iterate m initial particles by using the correlation optimization, particle individual extreme value, and particle swarm extreme value. Next, the specific configuration of the update and iteration module 40 will be described in detail. The update and iteration module 40 may further include: a clustering unit, configured to perform correlation normalization processing on the particle correlation, perform downward correlation ratio clustering starting from the highest correlation value, and establish a downward correlation ratio clustering result; a particle correlation unit, configured to end the clustering when the downward correlation ratio clustering result meets a preset threshold, and use the particles corresponding to the particle correlation in the clustering as correlation selection particles; a first iteration unit, configured to obtain the current iteration number and generate a first optimization weight based on the iteration number; a second iteration unit, configured to obtain the current iteration state and generate a second optimization weight based on the iteration state; an update unit, configured to perform comprehensive calculation on the first optimization weight and the second optimization weight, generate an optimization reward by using the low fitness particles in the correlation selection particles and the comprehensive calculation result, update the high fitness particles in the correlation selection particles by using the optimization reward, stop the update of the low fitness particles, and complete the update and iteration according to the update result.

[0038] Specifically, first, performing correlation normalization processing based on particle correlation data means performing correlation normalization processing. The particle correlation data may include, based on pulse waveform characteristic parameters (such as time domain and frequency domain characteristics), starting from the highest correlation value, gradually performing downward correlation ratio clustering. This means classifying the particle with the highest correlation value into one category, and then successively considering the particles with the second highest correlation value. If their correlation degree with the existing categories exceeds a certain threshold, they are classified into that category; otherwise, a new category is formed, thereby establishing a downward correlation ratio clustering result, that is, the particles included in each clustering and their correlation degree ratio, so as to establish the downward correlation ratio clustering result. Exemplarily, the preset threshold for the correlation ratio clustering result may be set to 70%, starting from the maximum for clustering, and stopping when 70% is satisfied. When the downward correlation ratio clustering result meets the preset threshold, the clustering ends. Using the particles corresponding to the particle correlation in the clustering as correlation selection particles means checking whether the clustering result meets the preset clustering quality threshold (such as the average correlation degree of particles within the clustering, the correlation degree difference between clusters, etc.). If the clustering result meets the preset clustering quality threshold, the clustering process ends, and the particle with the highest correlation degree is selected from the clustering result as the correlation selection particle. These particles will be used in the subsequent iteration optimization process.

[0039] Further record the current iteration number, and calculate the first optimization weight according to the iteration number and a preset weight adjustment strategy (such as linear decrease, exponential decay, etc.). This weight will be used in the subsequent particle update process to balance the global search and local search capabilities. At the same time, evaluate the current iteration state, which may include the distribution of particles, convergence speed, changes in the optimal solution, etc. Calculate the second optimization weight according to the iteration state and the preset weight adjustment strategy. This weight may be used to adjust the exploration ability of particles in the search space to cope with different situations during the iteration process.

[0040] Finally, comprehensively calculate the first optimization weight and the second optimization weight to obtain the comprehensive optimization weight. Among the associated selected particles, identify low-fitness particles and high-fitness particles. Generate an optimization reward according to the comprehensive optimization weight and the information of the low-fitness particles. The optimization reward can be an adjustment amount, a velocity vector, or new position coordinates. The optimization reward is used to guide the high-fitness particles to move towards a better solution. Update the high-fitness particles using the optimization reward, and at the same time stop the update of the low-fitness particles (or give them a smaller update step size). According to the update results, evaluate the new positions and fitness values of the particles, and determine whether the iteration termination conditions are met (such as reaching the maximum iteration number, the optimal solution remaining stable, etc.). If the conditions are met, the iteration process ends; otherwise, return to continue the iteration. By continuously iterating and improving the positions and associations of the particles, find the optimal PD detection scheme.

[0041] Next, the specific configuration of the update iteration module 40 will be described in detail. The update iteration module 40 may further include: a removal unit for removing the low-fitness particles that stop updating from the m initial particles to establish updated initial particles; a first individual update unit for performing individual updates on the non-associated selected particles in the updated initial particles based on the particle individual extreme value and the particle swarm extreme value to generate a first update result; a second individual update unit for performing individual updates on the associated selected particles in the updated initial particles based on the optimization reward, the particle individual extreme value, and the particle swarm extreme value to generate a second update result; a particle supplementation unit for supplementing new particles to the first update result and the second update result to complete the update iteration.

[0042] Specifically, removing the low-fitness particles that have stopped updating from the m initial particles can be achieved by monitoring the particle fitness, identifying the low-fitness particles, and removing the low-fitness particles. The process can be to continuously monitor the fitness value of each particle during the iteration of the hybrid particle swarm optimization algorithm. The fitness value is usually determined according to the objective function value of the particle, reflecting the quality of the particle in the solution space. At the same time, set a fitness threshold or determine the criteria for low-fitness particles according to the fitness distribution of the particles in the current iteration. Identify those particles with fitness lower than this criteria, and these particles are considered to be low-fitness particles that have stopped updating or fallen into local optima. Remove these low-fitness particles from the current m initial particles, leaving space for subsequent particle updates and supplements, thereby establishing updated initial particles.

[0043] Furthermore, perform individual updates on the non-associated selected particles in the updated initial particles based on the particle individual extreme value and the particle swarm extreme value. The non-associated selected particles in the updated initial particles refer to the part of the remaining particles that have not been selected as associated selected particles after removing the low-fitness particles. Then, for each non-associated selected particle, update its velocity and position according to its individual extreme value (i.e., the best position found by this particle so far) and the swarm extreme value (i.e., the best position found by the entire particle swarm so far). Based on this, obtain the first update result of the non-associated selected particles. The positions of these particles in the solution space have changed and may be closer to the global optimal solution.

[0044] Furthermore, perform individual updates on the associated selected particles in the updated initial particles based on the optimization reward, the particle individual extreme value, and the particle swarm extreme value. The associated selected particles in the updated initial particles refer to the particles that are selected as associated selected particles according to the downward association ratio clustering result and a preset threshold, which are closely related to the highest association value. For the associated selected particles, in addition to considering their individual extreme values and swarm extreme values, an optimization reward is introduced to further adjust their velocity and position. This optimization reward can be determined based on factors such as the contribution degree of the particle in the clustering and the improvement degree of the association value. Update the velocity and position of the associated selected particles by combining the individual extreme value, the swarm extreme value, and the optimization reward to obtain the second update result. The positions of these particles in the solution space can be more superior due to the introduction of the optimization reward.

[0045] Finally, adding new particles to the first update result and the second update result means determining the number of particles to be added according to the current number of the particle swarm and the preset particle swarm size. New particles are randomly generated in the solution space, and the positions and velocities of these particles can be initialized to random values or determined according to a certain heuristic strategy. The new particles are added to the first update result and the second update result to form an updated particle swarm. At the same time, the total number of the particle swarm is kept unchanged to maintain the search ability and diversity of the algorithm. Thus, a complete iteration process is completed. The updated particle swarm will be used as the input for the next iteration to continue the search and optimization process, so as to find the global optimal solution or an approximate optimal solution.

[0046] The denoising processing module 50 is used to dynamically adjust the inertia weight, perform iterative optimization, establish an optimal wavelet threshold according to the iterative optimization result, and perform denoising processing on the impulse signal set by using the optimal wavelet threshold to establish a denoised signal set.

[0047] Specifically, first, the linear decreasing weight algorithm is used to calculate the inertia weight:

[0048] ;

[0049] where is the inertia weight, is the maximum number of iterations, and are the minimum and maximum values of the weight coefficient respectively.

[0050] Furthermore, initial positions and velocities are randomly assigned to each particle. These positions represent possible wavelet threshold candidates. The fitness function value is calculated for the current position (i.e., the wavelet threshold candidate) of each particle. The fitness function can be defined based on the quality of the denoised signal (such as the signal-to-noise ratio SNR), and the inertia weight of each particle is dynamically adjusted according to the particle's historical performance (such as the improvement of the fitness value). For example, if a particle performs well in the past few iterations (the fitness value is significantly improved), the inertia weight is decreased to strengthen local search; conversely, if the performance is poor, the inertia weight is increased to promote global search.

[0051] Furthermore, the velocity and position of the particle are updated according to the current velocity, inertia weight, individual optimal position, and global optimal position of the particle. It is judged whether the maximum number of iterations is reached or the change in the fitness value is less than the preset threshold. If so, the iteration is stopped; otherwise, the next step is continued. Then, the current fitness value of each particle is compared with the fitness values of the individual optimal and global optimal. If it is better, the corresponding optimal position and fitness value are updated. When the iteration ends, the global optimal position is the required optimal wavelet threshold.

[0052] Further perform inverse wavelet transform on the processed wavelet coefficients to obtain a denoised signal set, and based on the optimal wavelet threshold determined above, perform threshold processing on the wavelet coefficients, that is, retain the coefficients greater than the threshold, set the coefficients less than or equal to the threshold to zero or perform shrinkage processing, and then perform inverse wavelet transform on the processed wavelet coefficients to obtain a denoised signal set.

[0053] Feature extraction module 60, which is used to extract first-order shape parameter features, kurtosis features, skewness features, and frequency domain features from the denoised signal set and establish feature extraction results; Next, as Figure 2 shown, the specific configuration of the feature extraction module 60 will be described in detail. The feature extraction module 60 may further include: a first feature calculation unit, which is used to calculate the kurtosis feature through the formula as follows:

[0054] ;

[0055] where represents kurtosis, represents the th value of the random variable, represents the mean of the random variable, is the standard deviation of the random variable, is the total number of random variables, represents the value of the probability density function of the random variable at ;

[0056] A second feature calculation unit, which is used to calculate the skewness feature through the formula as follows:

[0057] ;

[0058] where represents skewness;

[0059] A feature extraction unit, which is used to use the kurtosis feature and the skewness feature as the feature extraction results.

[0060] Specifically, as Figure 2 shown is the maximum amplitude of the discharge pulse; the pulse rise time is the time from 10% of the maximum amplitude to 90% of the maximum amplitude, that is: - ; the pulse fall time is the time from 90% of the maximum amplitude to 10% of the maximum amplitude, that is: - . The 50% amplitude pulse duration is the pulse waveform duration starting from 50% of the maximum amplitude at the rising edge to 50% of the maximum amplitude at the falling edge, that is: - ; The definition of the 10% maximum amplitude pulse duration is similar, - , thereby determining the first-order shape parameter characteristics.

[0061] Kurtosis describes the degree to which the probability distribution of a random variable is concentrated around the mean, or the rate of increase of the random variable, that is, the steepness of the change in the distribution function. Assume that the probability distribution function of the random variable X is f(x), the mean is µ, and the variance is , then

[0062] ;

[0063] where, represents kurtosis, represents the th value of the random variable, represents the mean of the random variable, is the standard deviation of the random variable, is the total number of random variables, represents the value of the probability density function of the random variable at ;

[0064] Treat each sampling point on the single-shot discharge waveform as a random variable, and use the sampling value of the partial discharge signal at each sampling moment and the probability of this value in this discharge to replace and respectively, then the above formula can be expressed as:

[0065] ;

[0066] The above formula is the kurtosis of the single-shot discharge waveform of the ultra-wideband partial discharge. The larger the , the greater the intensity of a discharge pulse in the single-shot discharge waveform. Kurtosis represents the "height, width, and shape" of the spectrogram. When is larger, it means that the data information is relatively sharp, that is, the distribution is concentrated. When

[0067] Skewness describes the symmetry of the probability distribution of a random variable. Take each sampling value on the time-domain signal waveform of the partial discharge as a random variable. Assume that the probability distribution function of the random variable is , the mean is µ, and the variance is σ2, then:

[0068] ;

[0069] Among them, characterizes skewness;

[0070] Similarly, by using a certain distribution parameter and its occurrence probability to replace and , the of this distribution can be obtained. It can be seen from the above formula that is the degree of symmetry of a certain distribution with respect to the mean value of the corresponding parameter of a single discharge waveform, or the deviation of the discharge distribution from the normal distribution. If >0, it means that its distribution is right-skewed with respect to the normal distribution. If <0, it means that its distribution is left-skewed with respect to the normal distribution. The asymmetry represents the difference in the discharge distribution between the positive and negative half-cycles, that is, the symmetry of the upper and lower amplitude distributions of a single partial discharge:

[0071] ;

[0072] Among them, N1 and N2 are the numbers of values greater than zero and less than zero of the sampling points of a single partial discharge respectively, and represent the values less than zero and greater than zero of the sampling points of a single partial discharge. It can be seen from the formula that >1 means that the average value of the positive half-cycle is greater than the average value of the negative half-cycle; =1 means complete symmetry up and down; <1 means that the average value of the positive half-cycle is less than the average value of the negative half-cycle. The pulse factor I and the margin factor L are both defined by conventional mathematics to obtain the skewness feature.

[0073] In addition to collecting the time-domain waveform data of the single discharge pulse of partial discharge, the waveform frequency-domain data is also collected. The waveform after denoising and removing redundant data is subjected to frequency-domain transformation using the fast Fourier transform, and the spectrogram is analyzed. The main concentration range of partial discharge and the frequency corresponding to the maximum amplitude are found from the transformed spectrogram; as the defect model changes, the frequency corresponding to the maximum amplitude also changes. According to the analysis of the amplitude-frequency characteristic diagrams of different discharge types of transformer oil-paper insulation, characteristic parameters are extracted from the amplitude-frequency characteristic diagrams, including: the frequency at the maximum amplitude, the average amplitude within the range of 0-30 MHz, and the average amplitude within the range of 30-50 MHz, so as to obtain the frequency-domain characteristics.

[0074] Integrate the first-order shape parameter characteristics, kurtosis characteristics, skewness characteristics, and frequency-domain characteristics obtained above to establish the feature extraction results.

[0075] Anomaly analysis module 70 is used to perform transformer anomaly analysis based on the feature extraction results and generate an anomaly report.

[0076] Next, the specific configuration of the anomaly analysis module 70 will be described in detail. The anomaly analysis module 70 may further include: a first parameter acquisition unit for acquiring equipment status parameters of the transformer and establishing an equipment status data set, where the equipment status data set includes load data, voltage level data, and operation time data; a second parameter acquisition unit for performing environmental parameter acquisition of the transformer and establishing an environmental data set, where the environmental data set includes temperature data and humidity data; an anomaly identification unit for using the equipment status data set, the environmental data set, and the feature extraction results to perform transformer anomaly identification and establish an anomaly report.

[0077] Specifically, the initial data parameters can be acquired through data acquisition devices (such as sensors, monitors, etc.) connected to the transformer, which can include load data, voltage level data, and operation time data. The load data is obtained by storing the load current, active power, reactive power, etc. of the transformer collected in real time through current transformers or power monitoring devices. The voltage level data is obtained by collecting the input and output voltage level data of the transformer using voltage transformers or voltage monitoring devices and recording and storing the obtained data. The operation time data is recorded by a time recording device (such as a clock module) or a data management system for the operation time of the transformer, including continuous operation time, cumulative operation time, etc. Finally, the load data, voltage level data, and operation time data are integrated to form an equipment status data set including load data, voltage level data, and operation time data.

[0078] Furthermore, based on the temperature sensors and humidity sensors deployed around the transformer, the status of the transformer operation environment can be accurately acquired, which can include temperature data and humidity data. The temperature data is obtained by storing the temperature data of the transformer and its surrounding environment collected in real time through temperature sensors. The humidity data is obtained by storing the humidity data of the environment collected in real time using humidity sensors. The temperature data and humidity data are integrated to form an environmental data set including temperature data and humidity data.

[0079] Finally, preprocess the device status dataset and the environmental dataset to extract key features that can reflect the operating status of the transformer, such as load change rate, voltage volatility, temperature anomaly points, humidity change trends, etc. Use machine learning algorithms such as support vector machines, decision trees, and neural networks to train the model with historical data including normal and abnormal status data so that it can identify the abnormal status of the transformer. Input the real-time collected device status data and environmental data into the trained anomaly recognition model, and the model will output the results of anomaly recognition according to the input data. When the model identifies an abnormal status, an anomaly report will be automatically generated. The report should include information such as the time of the anomaly occurrence, the type of anomaly (such as overload, overheat, high humidity, etc.), the degree of anomaly, and possible handling suggestions. According to the content of the anomaly report, take corresponding handling measures, such as adjusting the load, reducing the environmental temperature, increasing ventilation, etc., and feedback the handling results to the anomaly recognition model to optimize the recognition performance of the model, generate an anomaly report, and provide strong guarantee for the safe operation of the transformer.

[0080] Next, an ultra-wideband intelligent detection platform for transformer PD sources may further include: a focus monitoring module for establishing a focus window according to the anomaly report; a feedback monitoring module for performing feedback update on the feature extraction module based on the focus window and performing transformer monitoring management according to the feature extraction module after feedback update.

[0081] Specifically, the focus monitoring module first receives the anomaly report from the anomaly recognition unit. These reports contain information such as the time, type, degree of the transformer anomaly occurrence, and possible handling suggestions. According to the content of the anomaly report, the focus monitoring module will establish a focus window. The focus window can be a time interval, which is used to cover the time point of the anomaly occurrence and can extend a certain period of time forward and backward to comprehensively analyze the situation before and after the anomaly. Within the focus window, the focus monitoring module will set corresponding monitoring parameters according to the type of anomaly. Exemplarily, if the anomaly is caused by overload, then the monitoring parameters may include load current, active power, etc.; if the anomaly is caused by overheat, then the monitoring parameters may include temperature, heat dissipation situation, etc. After setting the monitoring parameters, the focus monitoring module will start real-time monitoring and continuously monitor the status of the transformer within the focus window.

[0082] Furthermore, the feedback monitoring module receives real-time monitoring data from the concerned monitoring module. These data are obtained by real-time monitoring of the transformer within the concerned window, and the feedback monitoring module will analyze the received real-time monitoring data. If the data indicates that the state of the transformer is still abnormal or the abnormal situation has worsened, the feedback mechanism will be triggered. After the feedback is triggered, the feedback monitoring module will update the feature extraction module according to the real-time monitoring data and the content of the abnormal report. The updated content may include adjusting the feature extraction algorithm, adding new feature dimensions, etc., so as to more accurately reflect the current state of the transformer. The updated feature extraction module will re-extract features from the real-time monitoring data. The extracted features will be used for subsequent abnormal identification and analysis. If the state of the transformer is still identified as abnormal after re-extracting features, a new abnormal report will be generated and sent to relevant personnel for processing.

[0083] Finally, the monitoring data collected by the concerned monitoring module and the feedback monitoring module are integrated to form a comprehensive data set. This data set contains various parameters and features of the transformer at different time periods and in different states. The integrated monitoring data is deeply analyzed and mined. Through data analysis, the laws and trends of the transformer state changes can be found, providing a scientific basis for subsequent monitoring and management. According to the results of data analysis, a more reasonable monitoring strategy can be formulated. For example, the monitoring frequency can be adjusted, more sensitive thresholds can be set, etc., so as to detect and handle the abnormal situation of the transformer more timely. According to the actual monitoring effect, the monitoring system is continuously optimized. This includes updating hardware devices, improving software algorithms, enhancing the stability and reliability of the system, etc. According to the formulated monitoring strategy and the optimized monitoring system, the transformer is continuously monitored and managed. Through means such as real-time monitoring, abnormal identification, and feedback update, the safe and stable operation of the transformer is ensured, thus realizing the comprehensive, accurate, and real-time monitoring and management of the transformer.

[0084] The embodiment of this application solves the technical problem of low accuracy in distinguishing ultra-wideband PD signal detection from noise sources, and achieves the technical effect of quickly extracting PD pulse characteristic parameters and realizing intelligent clustering and separation.

[0085] In the above text, with reference to Figure 1 a kind of intelligent detection platform for ultra-wideband PD sources of transformers according to the embodiment of this application is described in detail. Next, with reference to Figure 3 a kind of intelligent detection method for ultra-wideband PD sources of transformers according to the embodiment of this application will be described.

[0086] Step A100, configure a sensor group, the sensor group includes HFCT, bushing end screen sensors, and detection impedances, use the sensor group to monitor the partial discharge pulse signals of the transformer, and establish a pulse signal set;

[0087] Execute step A200. After performing wavelet decomposition on the pulse signal set, obtain the wavelet decomposition coefficients at each scale, and generate m initial particles in the solution space by using the wavelet decomposition coefficients.

[0088] Execute step A300. Calculate the particle similarity of the m initial particles, establish particle similarity associations, and perform particle combination by using the particle similarity associations to establish particle associations.

[0089] Configure the similarity calculation formula as follows:

[0090] ;

[0091] Where, represents the particle similarity association between particle and particle , and respectively represent the fitness functions for measuring the individual advantages and disadvantages of particle and particle , is the particle fitness deviation coefficient, and is the particle self-deviation coefficient;

[0092] Configure the combination isolation strategy. After sorting the association order of the particle similarity associations, perform association screening from the maximum association through the combination isolation strategy to complete particle combination and establish particle associations.

[0093] Execute step A400. After performing association selection on the particle associations, generate association optimization based on the current iteration number and iteration state, and update and iterate the m initial particles by using the association optimization, particle individual extreme values, and particle swarm extreme values.

[0094] After performing association normalization processing on the particle associations, perform downward association ratio clustering from the highest association value to establish a downward association ratio clustering result.

[0095] When the downward association ratio clustering result meets the preset threshold, the clustering ends, and the particles corresponding to the particle associations in the clustering are used as the association selection particles.

[0096] Obtain the current iteration number and generate a first optimization weight based on the iteration number.

[0097] Obtain the current iteration state and generate a second optimization weight based on the iteration state.

[0098] After comprehensively calculating the first optimization weight and the second optimization weight, use the low-fitness particles in the associated selection particles and the comprehensive calculation result to generate an optimization reward. Use the optimization reward to update the high-fitness particles in the associated selection particles, and stop the update of the low-fitness particles. Complete the update iteration according to the update result.

[0099] Remove the low-fitness particles whose update has stopped from the m initial particles, and establish updated initial particles;

[0100] Perform individual update on the non-associated selection particles in the updated initial particles based on the particle individual extreme value and the particle swarm extreme value to generate a first update result;

[0101] Perform individual update on the associated selection particles in the updated initial particles based on the optimization reward, the particle individual extreme value, and the particle swarm extreme value to generate a second update result;

[0102] After supplementing new particles to the first update result and the second update result, complete the update iteration.

[0103] Execute step A500, dynamically adjust the inertia weight, perform iterative optimization, establish an optimal wavelet threshold according to the iterative optimization result, and use the optimal wavelet threshold to perform denoising processing on the impulse signal set to establish a denoised signal set;

[0104] Next, execute step A600, extract the first-order shape parameter features, kurtosis features, skewness features, and frequency domain features from the denoised signal set to establish a feature extraction result;

[0105] Calculate the kurtosis feature through the formula as follows:

[0106] ;

[0107] Among them, represents kurtosis, represents the th value of the random variable, represents the mean value of the random variable, is the standard deviation of the random variable, is the total number of random variables, represents the value of the probability density function of the random variable at ;

[0108] Calculate the skewness feature through the formula as follows:

[0109] ;

[0110] Among them, represents skewness;

[0111] Take the kurtosis feature and skewness feature as the feature extraction results.

[0112] Finally, perform step A700 to analyze the abnormality of the transformer based on the feature extraction results and generate an abnormality report.

[0113] Collect the device status parameters of the transformer and establish a device status data set, where the device status data set includes load data, voltage level data, and operation time data;

[0114] Perform the collection of the environmental parameters of the transformer and establish an environmental data set, where the environmental data set includes temperature data and humidity data;

[0115] Use the device status data set, the environmental data set, and the feature extraction results to identify the abnormality of the transformer and establish an abnormality report.

[0116] The intelligent detection platform for the PD source of the ultra-wideband transformer further includes:

[0117] A focus monitoring module for establishing a focus window according to the abnormality report;

[0118] A feedback monitoring module for performing feedback update on the feature extraction module based on the focus window and performing transformer monitoring management according to the feature extraction module after feedback update.

[0119] The intelligent detection platform for the PD source of the ultra-wideband transformer provided by the embodiment of the present application can execute the intelligent detection method for the PD source of the ultra-wideband transformer provided by any embodiment of the present application, and has the corresponding functional modules and beneficial effects for executing the method.

[0120] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server. The included individual units and modules are only divided according to the functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present application.

[0121] The above specific implementation manners do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to the design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. An ultra-wideband transformer PD source intelligent detection platform, characterized in that: The platform includes: A configuration module is used to configure a sensor group, wherein the sensor group includes an HFCT, a bushing end screen sensor, and a detection impedance, and the sensor group is used to monitor the partial discharge pulse signal of the transformer to establish a pulse signal set; An initial solution construction module, used to perform wavelet decomposition on the pulse signal set, obtain wavelet decomposition coefficients at each scale, and generate m initial particles in the solution space using the wavelet decomposition coefficients; An association analysis module is used to calculate the particle similarity of the m initial particles, establish particle similarity association, combine particles using the particle similarity association, and establish particle association; An update iteration module is used to generate an association optimization through the current iteration number and iteration state after performing association selection on the particle association, and perform m initial particle update iterations using the association optimization and the particle individual extreme value and the particle group extreme value; A denoising processing module is used to dynamically adjust the inertia weight, perform iterative optimization, establish an optimal wavelet threshold according to the iterative optimization result, and establish a denoised signal set after denoising the pulse signal set using the optimal wavelet threshold; A feature extraction module, used to extract first-order shape parameter features, kurtosis features, skewness features, and frequency domain features from the denoised signal set, and establish a feature extraction result; The abnormality analysis module is used to perform transformer abnormality analysis based on the feature extraction results and generate an abnormality report.

2. The ultra-wideband transformer PD source intelligent detection platform according to claim 1, characterized in that: The association analysis module is used for: The first configuration unit is used to configure a similarity calculation formula as follows: ； in, Characterize the particle similarity correlation between particle A and particle B, and The fitness functions that measure the individual quality of particle B and particle A respectively, is the particle fitness deviation coefficient, is the particle's own deviation coefficient; The second configuration unit is used to configure a combination isolation strategy, after sorting the particle similar associations in association order, perform association screening from the maximum association through the combination isolation strategy to complete the particle combination and establish the particle association.

3. The ultra-wideband transformer PD source intelligent detection platform as claimed in claim 2, characterized in that: The update iteration module is used for: A clustering unit, configured to perform association normalization processing on the particle associations, and then perform downward association ratio clustering from the highest association value to establish a downward association ratio clustering result; The particle association unit is used to terminate clustering when the downward association ratio clustering result meets a preset threshold, and to associate the particles in the cluster with the particles corresponding to the particles as the association selection particles; A first iteration unit, used to obtain a current number of iterations, and generate a first optimization weight based on the number of iterations; Calculate the first optimization weight according to the number of iterations and the preset weight adjustment strategy; A second iteration unit, used to obtain a current iteration state, and generate a second optimization weight based on the iteration state; Calculate the second optimization weight according to the iteration state and the preset weight adjustment strategy; An updating unit is used to comprehensively calculate the first optimization weight and the second optimization weight, generate an optimization reward by using the low fitness particles in the associated selected particles and the comprehensive calculation result, update the high fitness particles in the associated selected particles by using the optimization reward, stop updating the low fitness particles, and complete the update iteration according to the update result.

4. The ultra-wideband transformer PD source intelligent detection platform as claimed in claim 3, characterized in that: The update iteration module is also used for: A removal unit is used to remove low-fitness particles that have stopped updating from the m initial particles, and establish updated initial particles; A first individual updating unit, used for performing individual updating on non-associated selected particles in the updated initial particles based on particle individual extreme values ​​and particle group extreme values, to generate a first updating result; A second individual updating unit is used to perform individual updating on the associated selected particles in the updated initial particles based on the optimization reward, the individual extreme value of the particles, and the extreme value of the particle group to generate a second update result; The particle supplement unit is used to supplement the first update result and the second update result with new particles to complete the update iteration.

5. The ultra-wideband transformer PD source intelligent detection platform as claimed in claim 1, characterized in that: The feature extraction module is used for: The first feature calculation unit is used to calculate the kurtosis feature by the formula as follows: ； in, Characterizes the kurtosis, Characterize the random variable A value, represents the mean of a random variable, is the standard deviation of the random variable, is the total number of random variables, The probability density function that characterizes a random variable is The value of ; The second feature calculation unit is used to calculate the skewness feature through the formula as follows: ； in, Characterize skewness; The feature extraction unit is used to take the kurtosis feature and the skewness feature as feature extraction results.

6. The ultra-wideband transformer PD source intelligent detection platform according to claim 1, characterized in that: The abnormality analysis module is also used for: A first parameter acquisition unit is used to collect equipment status parameters of the transformer and establish an equipment status data set, wherein the equipment status data set includes load data, voltage level data, and operating time data; A second parameter collection unit is used to collect environmental parameters of the transformer and establish an environmental data set, wherein the environmental data set includes temperature data and humidity data; The abnormality identification unit is used to identify transformer abnormalities using the equipment status data set, the environmental data set, and the feature extraction result, and to establish an abnormality report.

7. The ultra-wideband transformer PD source intelligent detection platform according to claim 1, characterized in that: The platform also includes: A focus monitoring module, used for establishing a focus window according to the abnormal report; A feedback monitoring module is used to perform feedback update on the feature extraction module based on the focus window, and perform transformer monitoring management according to the feature extraction module after feedback update.

8. An ultra-wideband transformer PD source intelligent detection method, characterized in that: The method is performed by an ultra-wideband transformer PD source intelligent detection platform according to any one of claims 1 to 7, wherein the ultra-wideband transformer PD source intelligent detection method comprises: A sensor group is configured, wherein the sensor group includes an HFCT, a bushing end screen sensor, and a detection impedance, and the sensor group is used to monitor the partial discharge pulse signal of the transformer to establish a pulse signal set; After performing wavelet decomposition on the pulse signal set, wavelet decomposition coefficients at each scale are obtained, and m initial particles are generated in the solution space using the wavelet decomposition coefficients; Calculating particle similarity for the m initial particles, establishing particle similarity association, combining particles using the particle similarity association, and establishing particle association; After the particle association is selected, an association optimization is generated according to the current iteration number and iteration state, and the association optimization and the particle individual extreme value and the particle group extreme value are used to perform m initial particle update iterations; Dynamically adjust the inertia weight, perform iterative optimization, establish an optimal wavelet threshold according to the iterative optimization result, and establish a denoised signal set after denoising the pulse signal set using the optimal wavelet threshold; Extracting first-order shape parameter features, kurtosis features, skewness features, and frequency domain features from the denoised signal set to establish a feature extraction result; Transformer abnormality analysis is performed based on the feature extraction results to generate an abnormality report.

Images