A converter transformer fault operation and maintenance method, system and terminal device

Abstract

The application provides a converter transformer fault operation and maintenance method, system and terminal equipment, which comprises the following steps: obtaining an operation state signal to decompose the operation state wavelet coefficient in a frequency band, reconstructing the operation state wavelet coefficient and the reconstructed operation state signal according to the wavelet coefficient; obtaining the operation state signal-to-noise ratio by intercepting the corresponding active section and noise section from the operation state signal and the reconstructed operation state signal, and obtaining the signal cumulative distance and the original signal cumulative fluctuation to generate the state waveform similarity; generating the state energy concentration degree according to the operation state wavelet coefficient and the reconstructed operation state wavelet coefficient; obtaining the low-frequency trend item according to the operation state signal and the reconstructed operation state signal to generate the state trend consistency degree; and obtaining the signal quality level based on the above-mentioned signal-to-noise ratio, waveform similarity, energy concentration degree and state trend consistency degree to screen the effective operation state signal for the converter transformer fault operation and maintenance. The application can improve the reliability of the converter transformer fault operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of fault operation and maintenance technology, and in particular to a fault operation and maintenance method, system and terminal equipment for converter transformers. Background Technology

[0002] High-voltage direct current (HVDC) transmission technology has been widely adopted, and the converter transformer, as the core equipment for AC / DC power conversion, directly impacts the safe and stable operation of the power grid. However, converter transformers typically operate in industrial environments characterized by strong electromagnetic interference, high mechanical stress, or complex thermo-coupling. Consequently, the operational status information collected for fault maintenance is susceptible to various noise interferences during acquisition, transmission, and processing, leading to signal distortion and severely affecting the reliability of fault maintenance for converter transformers.

[0003] To address the aforementioned issues, existing transformer fault maintenance solutions first construct a multi-scale wavelet transform and improved adaptive Kalman filter fusion architecture to acquire monitoring signals from the transformer windings. Multi-scale wavelet transform is used to perform frequency division processing on the monitoring signals to determine the coefficients for each scale. Then, the noise covariance matrix is ​​adjusted based on the improved adaptive Kalman filter, and the filtering results are determined by filtering the coefficients for each scale. Finally, the filtered results are reconstructed into a reconstructed signal for signal quality assessment to determine the signal quality of the transformer windings for transformer fault maintenance. However, converter transformers operate in highly turbulent industrial environments, resulting in significant signal distortion in their acquired operating status signals. While existing solutions filter out noise from the original signal, excessive smoothing can easily lead to the loss of useful transient components or edge information. Furthermore, in the presence of strong residual noise, low-quality signals may be mistakenly identified as valid signals, leading to misjudgments in converter transformer fault maintenance and significantly reducing the reliability of such maintenance. Summary of the Invention

[0004] The present invention aims to provide a method, system and terminal equipment for the fault operation and maintenance of converter transformers, so as to solve the above-mentioned technical problems and improve the reliability of fault operation and maintenance of converter transformers.

[0005] To address the aforementioned technical problems, this invention provides a method for fault operation and maintenance of converter transformers, comprising the following steps: Acquire the operating status signal of the target converter transformer; Based on the operating state signal, frequency band decomposition under a preset wavelet decomposition algorithm is used to obtain operating state wavelet coefficients, and based on the operating state wavelet coefficients, reconstructed operating state wavelet coefficients and reconstructed operating state signal are obtained under a preset signal optimization algorithm. The reconstructed running state active segment and the reconstructed running state noise segment are extracted from the reconstructed running state signal, and the original running state noise segment is extracted from the running state signal; The signal-to-noise ratio of the operating state is generated based on the reconstructed operating state activity segment, the reconstructed operating state noise segment, and the original operating state noise segment; Based on the operating status signal and the reconstructed operating status signal, the cumulative signal distance is obtained under a preset dynamic normalization distance algorithm, and the cumulative fluctuation of the original signal is obtained based on the operating status signal; The similarity of the generated state waveform is based on the signal accumulation distance and the original signal accumulation fluctuation; The state energy concentration is generated based on the wavelet coefficients of the operating state and the reconstructed operating state. Based on the operating status signal and the reconstructed operating status signal, the original low-frequency trend item and the reconstructed low-frequency trend item are obtained, and the state trend consistency degree is generated based on the original low-frequency trend item and the reconstructed low-frequency trend item. The signal quality level is obtained based on the operating state signal-to-noise ratio, the state waveform similarity, the state energy concentration, and the state trend consistency. The reconstructed operating status signals are filtered according to the signal quality level to obtain valid operating status signals, and the target converter transformer is subjected to fault maintenance based on the valid operating status signals.

[0006] The above-mentioned scheme recovers the reconstructed operating state wavelet coefficients and reconstructed operating state signal, which approximate the actual operating conditions of the converter transformer, from the operating state signal contaminated by strong interference noise. However, since the reconstructed signal and coefficients may still suffer from problems such as over-smoothing, loss of transient features, or noise residue, this scheme extracts the corresponding active and noise segments based on the reconstructed and original operating state signals to generate an operating state signal-to-noise ratio (SNR) that reflects the degree of interference suppression during the reconstruction process. This SNR can reduce the possibility of misjudging or missing subsequent transformer fault symptoms due to strong background noise interference. The higher the operating state SNR, the less noise contaminated the reconstructed signal is. However, the operating state SNR ignores the assessment of the similarity of the reconstructed signal morphology. Therefore, this scheme generates a state waveform similarity based on the signal accumulation distance and the accumulated fluctuation of the original signal to effectively verify whether the reconstruction process distorts or loses key waveform features such as transient impacts and oscillations related to the state changes of the converter transformer. If the state waveform similarity is low, it indicates that the reconstructed signal has lost the necessary features for fault identification. Physical morphology details; This scheme also generates a state energy concentration degree reflecting the distribution characteristics of signal energy based on the wavelet coefficients of the operating state and the reconstructed operating state, in order to identify broadband noise or artifacts remaining in the reconstructed signal. The higher the state energy concentration degree, the more effectively the irregular interference energy dispersed in each frequency band has been removed, and the signal characteristics are more focused. In addition, this scheme extracts the corresponding low-frequency trend terms from the operating state signal and the reconstructed operating state signal to generate a state trend consistency degree. Because it focuses on the low-frequency slow-change components of the signal, it can reflect the consistency of the trend information of the slow-change process of the transformer state before and after reconstruction, which complements the aforementioned signal-to-noise ratio, similarity and energy concentration degree that focus on instantaneous or local features. This scheme obtains a high reliability of signal quality level based on the above-mentioned operating state signal-to-noise ratio, state waveform similarity, state energy concentration and state trend consistency degree. Screening the reconstructed signal based on this level can improve the reliability of the obtained effective operating state signal. Using this effective operating state signal for fault operation and maintenance can significantly improve the reliability of converter transformer fault operation and maintenance.

[0007] Further, generating the operating state signal-to-noise ratio based on the reconstructed operating state active segment, the reconstructed operating state noise segment, and the original operating state noise segment includes: obtaining the effective signal power based on the reconstructed operating state active segment, and obtaining the noise distortion power based on the reconstructed operating state noise segment and the original operating state noise segment; generating the operating state signal-to-noise ratio based on the effective signal power and the noise distortion power.

[0008] In the above scheme, the effective signal power is first obtained based on the active segment of the reconstructed operating state. This effective signal power can reflect the energy intensity of the effective components retained in the active segment of the reconstructed operating state signal. The reconstructed operating state noise segment reflects the residual noise level of the reconstructed operating state signal after processing by the preset signal optimization algorithm, while the original operating state noise segment represents the inherent background noise level of the original operating state signal before processing. This scheme obtains the noise distortion power based on the reconstructed operating state noise segment and the original operating state noise segment, so that the noise distortion power can reflect the residual noise components of the reconstructed signal and also reflect the degree of change of the original signal noise distribution characteristics during the signal optimization process. Based on this, this scheme generates the signal-to-noise ratio of the operating status based on the effective signal power and noise distortion power, which can more accurately reflect the usability of the reconstructed operating status signal obtained after reconstruction optimization. This allows for more effective differentiation of low-quality signal segments with high noise distortion power and weak effective signal power when acquiring signal quality levels. Such low-quality signals are then filtered out of the effective operating status signal, ensuring that the effective operating status signal used for fault maintenance of the target converter transformer has higher reliability in terms of noise suppression. This reduces the risk of misjudging fault characteristics during fault maintenance due to residual noise or optimization distortion, and further improves the reliability of converter transformer fault maintenance.

[0009] Further, the step of obtaining the signal cumulative distance based on the operating status signal and the reconstructed operating status signal under a preset dynamic normalization distance algorithm, and obtaining the original signal cumulative fluctuation based on the operating status signal, includes: obtaining the signal cumulative distance based on the operating status signal and the reconstructed operating status signal under a preset dynamic normalization distance algorithm; obtaining the original signal length and the original signal standard deviation based on the operating status signal; and obtaining the original signal cumulative fluctuation based on the original signal length and the original signal standard deviation.

[0010] The above scheme uses a preset dynamic warping distance algorithm to nonlinearly align the running state signal and the reconstructed running state signal on the time axis. By matching the points one by one, the cumulative distance of the signals is obtained, which quantifies the overall difference in waveform morphology between the two signals. The scheme also obtains the original signal length and standard deviation based on the running state signal, and obtains the cumulative fluctuation of the original signal based on the original signal length and standard deviation to reflect the inherent fluctuation level of the original running state signal. This solution, based on the accumulated signal distance and the accumulated fluctuation of the original signal, generates a similarity of the state waveform. This adapts to the differences in the amplitude fluctuation range of the operating state signal under different operating conditions, avoiding deviations in the state waveform similarity assessment caused by the overall signal amplitude being too large or too small. It stably measures the fidelity of the waveform morphology of the operating state signal before and after reconstruction, effectively assessing whether the reconstruction process distorts or loses key transient and impact waveform features related to the equipment state. Consequently, in the subsequent process of acquiring signal quality levels and screening valid operating state signals, it can eliminate waveform distortion signals such as reconstructed operating state signals with distorted key transient waveform features. This ensures that valid operating state signals with good noise suppression and waveform fidelity are used for fault maintenance of the target converter transformer, reducing the risk of missed or misjudged fault features caused by signal morphology distortion, and further improving the reliability of converter transformer fault maintenance.

[0011] Further, the step of obtaining the original low-frequency trend item and the reconstructed low-frequency trend item based on the operating state signal and the reconstructed operating state signal, and generating a state trend consistency degree based on the original low-frequency trend item and the reconstructed low-frequency trend item, includes: obtaining the original low-frequency trend item and the reconstructed low-frequency trend item based on the operating state signal and the reconstructed operating state signal; obtaining the original low-frequency trend standard deviation and the reconstructed low-frequency trend standard deviation based on the original low-frequency trend item and the reconstructed low-frequency trend item, and obtaining the trend fluctuation degree based on the original low-frequency trend standard deviation and the reconstructed low-frequency trend standard deviation; obtaining the low-frequency trend correlation degree based on the original low-frequency trend item and the reconstructed low-frequency trend item; and generating a state trend consistency degree based on the low-frequency trend correlation degree and the trend fluctuation degree.

[0012] The above scheme first requires obtaining the original and reconstructed low-frequency trend terms based on the operating status signal and the reconstructed operating status signal. Then, the corresponding low-frequency trend standard deviation is obtained to determine the degree of trend fluctuation, quantifying the fluctuation amplitude relationship between the original and reconstructed low-frequency trend terms, reflecting the consistency of the low-frequency trend term fluctuation before and after reconstruction. This scheme also obtains the low-frequency trend correlation based on the original and reconstructed low-frequency trend terms to measure the strength of the correlation between the two low-frequency trend terms in terms of direction and degree of change. Subsequently, this scheme generates a state trend consistency degree based on this low-frequency trend correlation degree and the degree of trend fluctuation. This state trend consistency degree reflects the preservation of the low-frequency macroscopic change law between the original and reconstructed operating status signals, ensuring that the trend information reflecting the long-term slow degradation and temperature rise of the equipment is not destroyed. This consistency is then used to subsequently obtain signal quality levels and screen valid operating status signals. During the process, the reconstructed operating status signal with high waveform similarity but distorted low-frequency trend information due to non-proportional distortion of the low-frequency trend term amplitude characteristics can be eliminated. This ensures that the effective operating status signal used for fault maintenance of the target converter transformer has higher reliability in reflecting the long-term trend characteristics of the slow change process of equipment status. This enables the accurate capture of progressive fault signs caused by insulation aging, heat accumulation, or mechanical fatigue in the converter transformer, further reducing the risk of missed or false alarms of progressive faults due to low-frequency trend distortion and improving the reliability of converter transformer fault maintenance.

[0013] Further, the step of obtaining operating state wavelet coefficients through frequency band decomposition of the operating state signal using a preset wavelet decomposition algorithm, and obtaining reconstructed operating state wavelet coefficients and a reconstructed operating state signal based on the operating state wavelet coefficients using a preset signal optimization algorithm, includes: extracting key frequency band distribution features, signal stationarity features, and singularity distribution features from the operating state signal through time-frequency analysis; obtaining the optimal operating state wavelet basis by screening the key frequency band distribution features, the signal stationarity features, and the singularity distribution features under a preset wavelet function library and a preset tradeoff factor; and further reconstructing the operating state wavelet basis based on the operating state signal. The state signal is subjected to Fourier transform to obtain the state signal power spectrum, and the lowest effective signal frequency is obtained based on the state signal power spectrum; the optimal decomposition level is generated based on the lowest effective signal frequency under a preset sampling frequency and a preset safety factor; the state signal is decomposed into detail coefficients and approximation coefficients based on the optimal decomposition level and the optimal operating state wavelet basis, and the detail coefficients and approximation coefficients are integrated to obtain the operating state wavelet coefficients; the reconstructed operating state wavelet coefficients and the reconstructed operating state signal are obtained based on the operating state wavelet coefficients under a preset signal optimization algorithm.

[0014] Further, the step of obtaining the optimal operating state wavelet basis by screening based on the key frequency band distribution characteristics, the signal stationarity characteristics, and the singularity distribution characteristics under a preset wavelet function library and a preset trade-off factor includes: screening a candidate set of operating state wavelet bases based on the key frequency band distribution characteristics, the signal stationarity characteristics, and the singularity distribution characteristics in a preset wavelet function library; generating a signal matching degree index corresponding to each operating state wavelet base for each operating state wavelet base in the candidate set under the operating state signal and the preset trade-off factor; and screening the candidate set of operating state wavelet bases based on the signal matching degree index to obtain the optimal operating state wavelet basis.

[0015] The above scheme first extracts key frequency band distribution features, signal stationarity features, and singularity distribution features from the time-frequency analysis of the operating status signal. This reflects the time-frequency structure of the operating status signal from three dimensions: frequency distribution, stationarity, and abrupt change locations. This allows the waveform shape of the optimal operating status wavelet basis selected based on the above features to match the local features of the operating status signal. As a result, the optimal operating status wavelet coefficients more compactly express the effective components in the operating status signal, thus effectively separating noise components. Subsequently, this scheme obtains the power spectrum of the operating state signal based on the Fourier transform of the operating state signal, and obtains the lowest effective signal frequency based on the power spectrum. Then, based on this lowest frequency, it generates the optimal decomposition level under a preset sampling frequency and a preset safety factor, so that the optimal decomposition level is dynamically adjusted to adapt to the time-frequency characteristics of the operating state signal itself. Based on the optimal decomposition level and the optimal operating state wavelet basis, the operating state signal is decomposed into wavelet coefficients to obtain detail coefficients and approximation coefficients. Then, based on the operating state wavelet coefficients, the reconstructed operating state wavelet coefficients and the reconstructed operating state signal are obtained under a preset signal optimization algorithm. This ensures that the effective waveform features related to the target converter transformer state in the obtained reconstructed operating state signal can be effectively preserved, and noise components can be separated more effectively, improving the accuracy of the reconstructed operating state signal and reducing the risk of feature submersion or distortion due to wavelet basis mismatch. As a result, the subsequent operating state signal-to-noise ratio, state waveform similarity, state energy concentration, and state trend consistency are generated based on the highly accurate reconstructed operating state wavelet coefficients and the reconstructed operating state signal, improving the reliability of the obtained signal quality level, and thus improving the reliability of converter transformer fault operation and maintenance.

[0016] Further, obtaining the reconstructed operating state wavelet coefficients and the reconstructed operating state signal based on the operating state wavelet coefficients under a preset signal optimization algorithm includes: performing convolutional mapping on the detail coefficients of the operating state wavelet coefficients under the high-frequency path of a preset multi-path neural network to obtain a high-frequency deep feature representation; performing convolutional mapping on the approximate coefficients of the operating state wavelet coefficients under the low-frequency path of the preset multi-path neural network to obtain a low-frequency deep feature representation; performing convolutional mapping on the operating state signal under the full-band path of the preset multi-path neural network to obtain a full-band deep feature representation; and performing concatenation processing on the high-frequency deep feature representation, the low-frequency deep feature representation, and the full-band deep feature representation to obtain the reconstructed operating state wavelet coefficients and the reconstructed operating state signal.

[0017] The above scheme obtains a high-frequency deep feature representation by convolving the detail coefficients under the high-frequency path of a preset multi-path neural network. The detail coefficients contain components of the higher frequency band in the operating state signal. Through high-frequency path convolutional mapping, deep representations related to effective transient features can be extracted layer by layer from the detail coefficients, suppressing irregular random high-frequency interference. The scheme obtains a low-frequency deep feature representation by convolving the approximation coefficients under the low-frequency path of a preset multi-path neural network. The approximation coefficients carry the macroscopic trend information of the low-frequency band in the operating state signal, which reflects the long-term regularity of the slow change process of the target converter transformer state. Through low-frequency path convolutional mapping, the trend feature representation related to the slow change process of the equipment state can be maintained or even enhanced, avoiding the loss or distortion of low-frequency trend information. Furthermore, this scheme obtains a full-band deep feature representation by convolving the operating state signal under the full-band path of a preset multi-path neural network. Through this full-band path, global context information spanning different frequency bands can be obtained, effectively compensating for the information fragmentation that may occur when the high-frequency path and the low-frequency path focus on specific frequency bands. Therefore, the wavelet coefficients of the reconstructed operating state obtained by this scheme can achieve a balance between noise suppression, transient feature preservation and trend information, thereby reducing the risk of fault information distortion caused by feature loss or distortion during signal reconstruction and further improving the reliability of converter transformer fault operation and maintenance.

[0018] Further, after filtering the reconstructed operating status signals based on signal quality levels to obtain effective operating status signals, and performing fault maintenance on the target converter transformer based on the effective operating status signals, the method further includes: acquiring historical operating status signals, historical signal quality levels, and historical network parameter sets; extracting historical pure noise segments based on the historical operating status signals, and performing generalized Gaussian distribution fitting and maximum likelihood estimation based on the historical pure noise segments to obtain optimization trade-off factors and optimization safety factors; constructing an observation dataset based on the historical signal quality levels and the historical network parameter sets, and performing Bayesian optimization on the observation dataset under a preset maximum acquisition algorithm to obtain an optimized network parameter set; optimizing a preset multipath neural network based on the optimized network parameter set to obtain an optimized multipath neural network; and using the optimization trade-off factors, the optimization safety factors, and the optimized multipath neural network as the trade-off factors, safety factors, and multipath neural networks for the next execution of the converter transformer fault maintenance method.

[0019] In the above scheme, due to the long-term operation of the converter transformer, its operating conditions, on-site interference types, and noise statistical characteristics change slowly. If the preset relevant parameters remain unchanged, the adaptability of the subsequently reconstructed operating status signal to the actual working environment of the converter transformer will gradually decrease, thus affecting the selection quality of effective operating status signals. Therefore, this scheme, after fault maintenance, extracts historical pure noise segments based on the historical operating status signals, and obtains distribution parameters reflecting the current noise statistical characteristics by fitting and estimating the generalized Gaussian distribution of the historical pure noise segments. Based on this, the optimization trade-off factor and optimization safety factor can be obtained. Furthermore, this scheme also constructs an observation dataset based on the historical signal quality level and historical network parameter set, and obtains an optimized multi-path neural network through Bayesian optimization under the preset maximum acquisition algorithm, improving the reliability of the reconstructed operating status wavelet coefficients and reconstructed operating status signals obtained when this scheme is executed again. This scheme, based on the aforementioned optimized trade-off factors, optimized safety coefficients, and optimized multi-path neural networks, executes the next converter transformer fault operation and maintenance method. This enables the selection of the optimal operating state wavelet basis to better fit the current noise environment during subsequent execution, the determination of the optimal decomposition layer to be more reasonable, and the denoising and reconstruction performance of the preset multi-path neural network to be continuously improved. This, in turn, improves the reliability of the reconstructed operating state wavelet coefficients and reconstructed operating state signals, ensuring that the subsequently generated operating state signal-to-noise ratio, state waveform similarity, state energy concentration, and state trend consistency are generated on the basis of highly reliable reconstructed operating state wavelet coefficients and reconstructed operating state signals. This guarantees the reliability of the obtained signal quality level and reduces the risk of decreased fault operation and maintenance reliability due to adaptive attenuation.

[0020] This invention also provides a converter transformer fault operation and maintenance system for implementing any of the converter transformer fault operation and maintenance methods described above, comprising: an operation status signal acquisition module for acquiring the operation status signal of the target converter transformer; a signal and wavelet coefficient reconstruction module for obtaining operation status wavelet coefficients by frequency band decomposition of the operation status signal under a preset wavelet decomposition algorithm, and obtaining reconstructed operation status wavelet coefficients and a reconstructed operation status signal based on the operation status wavelet coefficients under a preset signal optimization algorithm; an active segment noise segment extraction module for extracting reconstructed operation status active segments and reconstructed operation status noise segments from the reconstructed operation status signal, and extracting original operation status noise segments from the operation status signal; a status signal-to-noise ratio generation module for generating an operation status signal-to-noise ratio based on the reconstructed operation status active segments, the reconstructed operation status noise segments, and the original operation status noise segments; and an accumulated fluctuation and distance generation module for generating a cumulative fluctuation and distance based on the operation status signal and the reconstructed operation status signal at a preset dynamic normalization distance. The algorithm acquires the cumulative distance of the signal and obtains the cumulative fluctuation of the original signal based on the operating state signal; a waveform similarity generation module is used to generate state waveform similarity based on the cumulative distance of the signal and the cumulative fluctuation of the original signal; an energy concentration generation module is used to generate state energy concentration based on the wavelet coefficients of the operating state and the reconstructed operating state; a trend consistency generation module is used to obtain the original low-frequency trend term and the reconstructed low-frequency trend term based on the operating state signal and the reconstructed operating state signal, and generate state trend consistency based on the original low-frequency trend term and the reconstructed low-frequency trend term; a signal quality level acquisition module is used to obtain the signal quality level based on the operating state signal-to-noise ratio, the state waveform similarity, the state energy concentration, and the state trend consistency; a fault operation and maintenance module is used to filter the reconstructed operating state signal according to the signal quality level to obtain a valid operating state signal, and perform fault operation and maintenance on the target converter transformer based on the valid operating state signal.

[0021] The present invention also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements any of the above-mentioned converter transformer fault operation and maintenance methods.

[0022] The above scheme obtains the reconstructed operating state wavelet coefficients and the reconstructed operating state signal from the operating state signal contaminated by strong interference noise. Based on the reconstructed and original operating state signals, it extracts corresponding active and noise segments to generate the operating state signal-to-noise ratio (SNR). This SNR can reduce the possibility of misjudging or missing subsequent transformer fault symptoms due to strong background noise interference. A higher SNR indicates less noise contamination of the reconstructed signal. However, the operating state SNR ignores the assessment of the similarity of the reconstructed signal's morphology. Therefore, this scheme generates state waveform similarity based on the cumulative signal distance and the cumulative fluctuation of the original signal to effectively verify whether the reconstruction process distorts or loses key waveform features related to the converter transformer's state changes, such as transient impacts and oscillations. If the state waveform similarity is low, it indicates that the reconstructed signal has lost the physical morphological details required for fault identification. This scheme also generates state waveform similarity based on the operating state wavelet coefficients and the reconstructed operating state wavelet coefficients. Energy concentration is used to identify broadband noise or artifacts remaining in the reconstructed signal. The higher the energy concentration, the more effectively the irregular interference energy scattered across various frequency bands has been removed, and the more focused the signal characteristics. Furthermore, this scheme extracts the corresponding low-frequency trend terms from the operating state signal and the reconstructed operating state signal to generate a state trend consistency score. Because this score focuses on the low-frequency slow-change components of the signal, it reflects the consistency of the trend information of the transformer's slow-change process before and after signal reconstruction, complementing the aforementioned signal-to-noise ratio, similarity, and energy concentration scores that focus on instantaneous or local features. The signal quality level obtained by this scheme based on the aforementioned operating state signal-to-noise ratio, state waveform similarity, state energy concentration, and state trend consistency score has high reliability. Screening the reconstructed signal based on this level can improve the reliability of the obtained effective operating state signal. Therefore, using this effective operating state signal for fault maintenance can significantly improve the reliability of converter transformer fault maintenance. Attached Figure Description

[0023] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating the technical implementation of a converter transformer fault operation and maintenance method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a converter transformer fault operation and maintenance system architecture provided in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0027] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0028] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0029] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0030] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0031] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0032] Please see Figure 1 This embodiment provides a method for fault operation and maintenance of converter transformers, including the following steps: Step S1: Obtain the operating status signal of the target converter transformer; Step S2: Based on the running state signal, frequency band decomposition is performed under a preset wavelet decomposition algorithm to obtain running state wavelet coefficients, and based on the running state wavelet coefficients, reconstructed running state wavelet coefficients and reconstructed running state signal are obtained under a preset signal optimization algorithm. Step S3: Extract the reconstructed running state active segment and the reconstructed running state noise segment from the reconstructed running state signal, and extract the original running state noise segment from the running state signal; Step S4: Generate the signal-to-noise ratio of the operating state based on the reconstructed operating state activity segment, the reconstructed operating state noise segment, and the original operating state noise segment; Step S5: Based on the operating status signal and the reconstructed operating status signal, obtain the signal cumulative distance under the preset dynamic normalization distance algorithm, and obtain the original signal cumulative fluctuation based on the operating status signal; Step S6: Generate state waveform similarity based on the signal accumulation distance and the original signal accumulation fluctuation; Step S7: Generate state energy concentration based on the wavelet coefficients of the operating state and the reconstructed operating state; Step S8: Obtain the original low-frequency trend item and the reconstructed low-frequency trend item based on the operating status signal and the reconstructed operating status signal, and generate the state trend consistency degree based on the original low-frequency trend item and the reconstructed low-frequency trend item; Step S9: Obtain the signal quality level based on the operating state signal-to-noise ratio, the state waveform similarity, the state energy concentration, and the state trend consistency. Step S10: Filter the reconstructed operating status signal according to the signal quality level to obtain a valid operating status signal, and perform fault maintenance on the target converter transformer according to the valid operating status signal.

[0033] After recovering the reconstructed operating state wavelet coefficients and reconstructed operating state signal that approximate the actual operating condition of the converter transformer from the operating state signal contaminated by strong interference noise in the above embodiments, since the reconstructed signal and coefficients may still have problems such as over-smoothing, loss of transient features, or noise residue, this embodiment extracts the corresponding active and noise segments based on the reconstructed operating state signal and the original operating state signal to generate an operating state signal-to-noise ratio that reflects the degree of suppression of interference during the reconstruction process. Based on this signal-to-noise ratio, the situation of misjudging or missing subsequent transformer fault symptoms caused by strong background noise interference can be reduced. The higher the operating state signal-to-noise ratio, the less the reconstructed signal is contaminated by noise. However, the operating state signal-to-noise ratio ignores the evaluation of the similarity of the reconstructed signal shape. Therefore, this embodiment generates a state waveform similarity based on the signal cumulative distance and the cumulative fluctuation of the original signal to effectively verify whether the reconstruction process distorts or loses key waveform features such as transient impacts and oscillations related to the state changes of the converter transformer. If the state waveform similarity is low, it indicates that the reconstructed signal has lost the necessary features for fault identification. Physical morphology details; This embodiment also generates a state energy concentration degree reflecting the distribution characteristics of signal energy based on the wavelet coefficients of the operating state and the reconstructed operating state, in order to identify broadband noise or artifacts remaining in the reconstructed signal. The higher the state energy concentration degree, the more effectively the irregular interference energy dispersed in each frequency band has been removed, and the signal characteristics are more focused. In addition, this embodiment extracts the corresponding low-frequency trend terms from the operating state signal and the reconstructed operating state signal to generate a state trend consistency degree. Because it focuses on the low-frequency slow-change component of the signal, it can reflect the consistency of the trend information of the slow-change process of the transformer state before and after reconstruction, which complements the aforementioned signal-to-noise ratio, similarity and energy concentration degree that focus on instantaneous or local features. This embodiment obtains a high reliability of signal quality level based on the above-mentioned operating state signal-to-noise ratio, state waveform similarity, state energy concentration and state trend consistency degree. Screening the reconstructed signal based on this level can improve the reliability of the obtained effective operating state signal. Using the effective operating state signal for fault operation and maintenance can significantly improve the reliability of converter transformer fault operation and maintenance.

[0034] Further, generating the operating state signal-to-noise ratio based on the reconstructed operating state active segment, the reconstructed operating state noise segment, and the original operating state noise segment includes: obtaining the effective signal power based on the reconstructed operating state active segment, and obtaining the noise distortion power based on the reconstructed operating state noise segment and the original operating state noise segment; generating the operating state signal-to-noise ratio based on the effective signal power and the noise distortion power.

[0035] In the above embodiments, the effective signal power is first obtained based on the active segment of the reconstructed operating state. This effective signal power can reflect the energy intensity of the effective components retained in the active segment of the reconstructed operating state signal. The reconstructed operating state noise segment reflects the residual noise level of the reconstructed operating state signal after processing by a preset signal optimization algorithm, while the original operating state noise segment represents the inherent background noise level of the original operating state signal before processing. In this embodiment, the noise distortion power is obtained based on the reconstructed operating state noise segment and the original operating state noise segment, so that the noise distortion power can reflect the residual noise components of the reconstructed signal and also reflect the degree of change in the noise distribution characteristics of the original signal during the signal optimization process. Based on this, this embodiment generates the signal-to-noise ratio of the operating status based on the effective signal power and noise distortion power, which can more accurately reflect the usability of the reconstructed operating status signal obtained after reconstruction optimization. This allows for more effective differentiation of low-quality signal segments with high noise distortion power and low effective signal power when acquiring signal quality levels. Such low-quality signals are filtered out of the effective operating status signal, ensuring that the effective operating status signal used for fault maintenance of the target converter transformer has higher reliability in terms of noise suppression. This reduces the risk of misjudging fault characteristics during fault maintenance due to residual noise or optimization distortion, and further improves the reliability of converter transformer fault maintenance.

[0036] Further, the step of obtaining the signal cumulative distance based on the operating status signal and the reconstructed operating status signal under a preset dynamic normalization distance algorithm, and obtaining the original signal cumulative fluctuation based on the operating status signal, includes: obtaining the signal cumulative distance based on the operating status signal and the reconstructed operating status signal under a preset dynamic normalization distance algorithm; obtaining the original signal length and the original signal standard deviation based on the operating status signal; and obtaining the original signal cumulative fluctuation based on the original signal length and the original signal standard deviation.

[0037] In the above embodiments, the preset dynamic warping distance algorithm can non-linearly align the running state signal and the reconstructed running state signal on the time axis, obtain the cumulative distance of the signal by point-by-point matching, and quantify the overall difference in waveform morphology between the two signals. In this embodiment, the original signal length and the original signal standard deviation are obtained based on the running state signal, and the cumulative fluctuation of the original signal is obtained based on the original signal length and the original signal standard deviation to reflect the inherent fluctuation level of the original running state signal. This embodiment subsequently relies on the state waveform similarity generated based on the aforementioned signal accumulation distance and the original signal accumulation fluctuation. This allows it to adapt to the differences in the amplitude fluctuation range of the operating state signal under different operating conditions, avoiding deviations in the state waveform similarity assessment caused by the overall signal amplitude being too large or too small. This ensures a stable measurement of the fidelity of the waveform morphology of the operating state signal before and after reconstruction, effectively assessing whether the reconstruction process distorts or loses key transient and impact waveform features related to the equipment state. Consequently, in the subsequent process of acquiring signal quality levels and screening valid operating state signals, it can eliminate waveform distortion signals such as reconstructed operating state signals with distorted key transient waveform features. This ensures that valid operating state signals with good noise suppression and waveform fidelity are used for fault maintenance of the target converter transformer, reducing the risk of missed or misjudged fault features caused by signal morphology distortion, and further improving the reliability of converter transformer fault maintenance.

[0038] Further, the step of obtaining the original low-frequency trend item and the reconstructed low-frequency trend item based on the operating state signal and the reconstructed operating state signal, and generating a state trend consistency degree based on the original low-frequency trend item and the reconstructed low-frequency trend item, includes: obtaining the original low-frequency trend item and the reconstructed low-frequency trend item based on the operating state signal and the reconstructed operating state signal; obtaining the original low-frequency trend standard deviation and the reconstructed low-frequency trend standard deviation based on the original low-frequency trend item and the reconstructed low-frequency trend item, and obtaining the trend fluctuation degree based on the original low-frequency trend standard deviation and the reconstructed low-frequency trend standard deviation; obtaining the low-frequency trend correlation degree based on the original low-frequency trend item and the reconstructed low-frequency trend item; and generating a state trend consistency degree based on the low-frequency trend correlation degree and the trend fluctuation degree.

[0039] The above embodiment first obtains the original low-frequency trend term and the reconstructed low-frequency trend term based on the operating status signal and the reconstructed operating status signal. Then, it obtains the corresponding low-frequency trend standard deviation to obtain the trend fluctuation degree, so as to quantify the fluctuation amplitude relationship between the original low-frequency trend term and the reconstructed low-frequency trend term, which can reflect the consistency of the low-frequency trend term fluctuation degree before and after reconstruction. This embodiment also obtains the low-frequency trend correlation degree based on the original low-frequency trend term and the reconstructed low-frequency trend term, which is used to measure the strength of the correlation between the two low-frequency trend terms in terms of the direction and degree of change. Subsequently, this embodiment generates the state trend consistency degree based on the low-frequency trend correlation degree and the trend fluctuation degree, so that the state trend consistency degree can reflect the preservation of the low-frequency macroscopic change law between the original operating status signal and the reconstructed operating status signal, and ensure that the trend information reflecting the long-term slow degradation, temperature rise and other processes of the equipment is not destroyed. This is then used to obtain the signal quality level and screen the effective operating status signal. During the process, the reconstructed operating status signal with high waveform similarity but distorted low-frequency trend information due to non-proportional distortion of the low-frequency trend term amplitude characteristics can be eliminated. This ensures that the effective operating status signal used for fault maintenance of the target converter transformer has higher reliability in reflecting the long-term trend characteristics of the slow change process of equipment status. This enables the accurate capture of progressive fault signs caused by insulation aging, heat accumulation, or mechanical fatigue in the converter transformer, further reducing the risk of missed or false alarms of progressive faults due to low-frequency trend distortion and improving the reliability of converter transformer fault maintenance.

[0040] Further, the step of obtaining operating state wavelet coefficients through frequency band decomposition of the operating state signal using a preset wavelet decomposition algorithm, and obtaining reconstructed operating state wavelet coefficients and a reconstructed operating state signal based on the operating state wavelet coefficients using a preset signal optimization algorithm, includes: extracting key frequency band distribution features, signal stationarity features, and singularity distribution features from the operating state signal through time-frequency analysis; obtaining the optimal operating state wavelet basis by screening the key frequency band distribution features, the signal stationarity features, and the singularity distribution features under a preset wavelet function library and a preset tradeoff factor; and further reconstructing the operating state wavelet basis based on the operating state signal. The state signal is subjected to Fourier transform to obtain the state signal power spectrum, and the lowest effective signal frequency is obtained based on the state signal power spectrum; the optimal decomposition level is generated based on the lowest effective signal frequency under a preset sampling frequency and a preset safety factor; the state signal is decomposed into detail coefficients and approximation coefficients based on the optimal decomposition level and the optimal operating state wavelet basis, and the detail coefficients and approximation coefficients are integrated to obtain the operating state wavelet coefficients; the reconstructed operating state wavelet coefficients and the reconstructed operating state signal are obtained based on the operating state wavelet coefficients under a preset signal optimization algorithm.

[0041] Further, the step of obtaining the optimal operating state wavelet basis by screening based on the key frequency band distribution characteristics, the signal stationarity characteristics, and the singularity distribution characteristics under a preset wavelet function library and a preset trade-off factor includes: screening a candidate set of operating state wavelet bases based on the key frequency band distribution characteristics, the signal stationarity characteristics, and the singularity distribution characteristics in a preset wavelet function library; generating a signal matching degree index corresponding to each operating state wavelet base for each operating state wavelet base in the candidate set under the operating state signal and the preset trade-off factor; and screening the candidate set of operating state wavelet bases based on the signal matching degree index to obtain the optimal operating state wavelet basis.

[0042] The above embodiments first extract key frequency band distribution features, signal stationarity features, and singularity distribution features from the time-frequency analysis of the operating status signal. This reflects the time-frequency structure of the operating status signal from three dimensions: frequency distribution, stationarity, and abrupt change locations. This allows the waveform shape of the optimal operating status wavelet basis selected based on the above features to match the local features of the operating status signal. As a result, the optimal operating status wavelet coefficients more compactly express the effective components in the operating status signal, thus effectively separating noise components. Subsequently, this embodiment obtains the power spectrum of the operating state signal based on the Fourier transform of the operating state signal, and obtains the lowest frequency of the effective signal based on the power spectrum of the operating state signal. Then, based on the lowest frequency, the optimal decomposition level is generated under a preset sampling frequency and a preset safety factor, so that the optimal decomposition level is adapted to the dynamic adjustment of the time-frequency characteristics of the operating state signal itself. Based on the optimal decomposition level and the optimal operating state wavelet basis, the operating state signal is decomposed by wavelet to obtain detail coefficients and approximation coefficients. Then, based on the operating state wavelet coefficients, the reconstructed operating state wavelet coefficients and the reconstructed operating state signal are obtained under a preset signal optimization algorithm. This ensures that the effective waveform features related to the target converter transformer state in the obtained reconstructed operating state signal can be effectively preserved, and noise components can be separated more effectively, improving the accuracy of the reconstructed operating state signal and reducing the risk of feature submersion or distortion due to wavelet basis mismatch. As a result, the subsequent operating state signal-to-noise ratio, state waveform similarity, state energy concentration, and state trend consistency are generated on the basis of the highly accurate reconstructed operating state wavelet coefficients and the reconstructed operating state signal, improving the reliability of the obtained signal quality level, and thus improving the reliability of converter transformer fault operation and maintenance.

[0043] Further, obtaining the reconstructed operating state wavelet coefficients and the reconstructed operating state signal based on the operating state wavelet coefficients under a preset signal optimization algorithm includes: performing convolutional mapping on the detail coefficients of the operating state wavelet coefficients under the high-frequency path of a preset multi-path neural network to obtain a high-frequency deep feature representation; performing convolutional mapping on the approximate coefficients of the operating state wavelet coefficients under the low-frequency path of the preset multi-path neural network to obtain a low-frequency deep feature representation; performing convolutional mapping on the operating state signal under the full-band path of the preset multi-path neural network to obtain a full-band deep feature representation; and performing concatenation processing on the high-frequency deep feature representation, the low-frequency deep feature representation, and the full-band deep feature representation to obtain the reconstructed operating state wavelet coefficients and the reconstructed operating state signal.

[0044] The above embodiments obtain high-frequency deep feature representations by convolving detail coefficients on the high-frequency path of a preset multi-path neural network. These detail coefficients contain components of higher frequency bands in the operating status signal. Through high-frequency path convolutional mapping, deep representations related to effective transient features can be extracted layer by layer from the detail coefficients, suppressing irregular random high-frequency interference. Approximation coefficients are obtained by convolving on the low-frequency path of a preset multi-path neural network. These approximation coefficients carry macroscopic trend information of the low-frequency band in the operating status signal, reflecting the long-term regularity of the slow change process of the target converter transformer's state. Through low-frequency path convolutional mapping, trend feature representations related to the slow change process of the equipment state can be maintained or even enhanced, avoiding the loss or distortion of low-frequency trend information. Furthermore, this embodiment obtains full-band deep feature representations by convolving the operating status signal on the full-band path of a preset multi-path neural network. Through this full-band path, global context information spanning different frequency bands can be obtained, effectively compensating for the information fragmentation that may occur when high-frequency and low-frequency paths focus on specific frequency bands. Therefore, the reconstructed operating state wavelet coefficients and the reconstructed operating state signal obtained in this embodiment can achieve a balance between noise suppression, transient feature preservation and trend information, thereby reducing the risk of fault information distortion caused by feature loss or distortion during signal reconstruction, and further improving the reliability of converter transformer fault operation and maintenance.

[0045] Further, after filtering the reconstructed operating status signals based on signal quality levels to obtain effective operating status signals, and performing fault maintenance on the target converter transformer based on the effective operating status signals, the method further includes: acquiring historical operating status signals, historical signal quality levels, and historical network parameter sets; extracting historical pure noise segments based on the historical operating status signals, and performing generalized Gaussian distribution fitting and maximum likelihood estimation based on the historical pure noise segments to obtain optimization trade-off factors and optimization safety factors; constructing an observation dataset based on the historical signal quality levels and the historical network parameter sets, and performing Bayesian optimization on the observation dataset under a preset maximum acquisition algorithm to obtain an optimized network parameter set; optimizing a preset multipath neural network based on the optimized network parameter set to obtain an optimized multipath neural network; and using the optimization trade-off factors, the optimization safety factors, and the optimized multipath neural network as the trade-off factors, safety factors, and multipath neural networks for the next execution of the converter transformer fault maintenance method.

[0046] In the above embodiments, since the long-term operation of the converter transformer causes its operating conditions, field interference types, and noise statistical characteristics to change slowly, if the preset relevant parameters remain unchanged, the adaptability of the subsequent reconstructed operating status signal to the actual working environment of the converter transformer will gradually decrease, thereby affecting the selection quality of effective operating status signals. Therefore, in this embodiment, after fault maintenance, a historical pure noise segment is extracted based on the historical operating status signal, and the distribution parameters reflecting the current noise statistical characteristics can be effectively obtained by fitting the generalized Gaussian distribution of the historical pure noise segment and estimating the maximum likelihood. Based on this, the optimization trade-off factor and optimization safety factor can be obtained. Furthermore, this embodiment also constructs an observation dataset based on the historical signal quality level and historical network parameter set, and obtains an optimized multi-path neural network through Bayesian optimization under the preset maximum acquisition algorithm, thereby improving the reliability of the reconstructed operating status wavelet coefficients and reconstructed operating status signals obtained when this embodiment is executed again. This embodiment, based on the aforementioned optimized trade-off factors, optimized safety coefficients, and optimized multi-path neural networks, executes the next converter transformer fault maintenance method. This enables the selection of the optimal operating state wavelet basis to better fit the current noise environment during subsequent executions, the determination of the optimal decomposition layer to be more reasonable, and the denoising and reconstruction performance of the preset multi-path neural network to be continuously improved. This, in turn, enhances the reliability of the reconstructed operating state wavelet coefficients and the reconstructed operating state signal. Consequently, the subsequently generated operating state signal-to-noise ratio, state waveform similarity, state energy concentration, and state trend consistency are generated based on the highly reliable reconstructed operating state wavelet coefficients and the reconstructed operating state signal, ensuring the reliability of the obtained signal quality level and reducing the risk of decreased fault maintenance reliability due to adaptive attenuation.

[0047] In one embodiment, the target converter transformer acquires raw time-domain signals such as vibration signals, current signals, and oil temperature signals through an online monitoring system. Then, null segments caused by communication interruptions are removed from the raw time-domain signals, and obvious abnormal transition points are corrected. The processed raw time-domain signals are then used as the operating status signals. ; Operating status signals Key frequency band distribution features, signal stationarity features, and singularity distribution features are extracted using time-frequency analysis techniques such as short-time Fourier transform or wavelet packet analysis. Based on these key frequency band distribution features, signal stationarity features, and singularity distribution features, several wavelet series that match the signal characteristics in terms of support length, vanishing moment, and symmetry are selected from a typical wavelet function library. These series include Daubechies, Symlets, and Coiflets, forming a candidate set of wavelet basis for operational states. ,in, Let A be a wavelet basis in the candidate set of wavelet basis for running states, and let the formula for constructing the candidate set of wavelet basis for running states be as follows: ; in, To obtain the running status signal Extracted feature vectors such as the frequency band where the main energy is concentrated, the zero-crossing rate of the signal, and the amplitude distribution skewness; To measure the wavelet basis of operation With feature vectors A function for the fit between them; This is the preset adaptation threshold.

[0048] Subsequently, the candidate set of wavelet bases for the running state was analyzed. For each wavelet basis of the operating state, in the operating state signal With preset trade-off factors Below, a signal matching degree index is generated corresponding to each operating state wavelet basis, and an operating state wavelet basis... Generate corresponding signal matching index The relevant formulas are as follows: ; Among them, the pre-set trade-off factors Set to 0.6; A function for calculating the energy percentage of the top 10% with the highest coefficients; A function to reconstruct the root mean square error. From the wavelet basis candidate set in the running state. In the process, the matching degree index of the selected signal is used. The wavelet basis with the largest operating state is used as the optimal operating state wavelet basis. .

[0049] Based on operating status signals The power spectrum of the state signal is obtained by performing a fast Fourier transform to identify the operating state signal. Lowest frequency of effective signal Then according to the preset sampling frequency With preset safety factor Generate the optimal number of decomposition layers The relevant formulas are as follows: ; Among them, the preset safety factor Set to 1.3.

[0050] Based on the optimal decomposition level and optimal operating state wavelet basis Operating status signals conduct Layer wavelet decomposition yields detail coefficients and approximation coefficients The detail coefficients and approximation coefficients are then integrated into the wavelet coefficients of the running state; subsequently, the detail coefficients of the wavelet coefficients of the running state are... High-frequency deep feature representations are obtained through convolutional mapping processing on the high-frequency paths of a pre-defined multi-path neural network; the approximate coefficients of the wavelet coefficients in the running state are then used. Low-frequency deep feature representations are obtained by convolutional mapping under the low-frequency path of the preset multi-path neural network; the running state signal is then processed. Full-band deep feature representations are obtained through convolutional mapping processing along the full-band paths of a pre-defined multi-path neural network; each path consists of L concatenated residual blocks, and each residual block performs the following operation: ; in, Input for the residual block; Output for residual blocks; W i} represents the set of learnable convolutional kernel weight parameters within the residual block, with subscripts... i Indicates the first... i Each residual block typically contains two convolutional layers, corresponding to i=1,2. The set of weight parameters of the convolutional kernel is iteratively updated during network training through backpropagation and gradient descent algorithms to adaptively extract effective patterns from the input features. It is a mapping function that includes convolution, batch normalization, and ReLU activation.

[0051] It should be noted that after the input of each path undergoes a layer-by-layer nonlinear transformation through all L residual blocks of that path, its output is the deep feature representation learned by that path. This deep feature representation can be recorded as the intermediate feature map of the path, where This refers to the number of characteristic channels, which correspond to different frequency bands or characteristic modes. The length of the feature sequence.

[0052] In one embodiment, a channel-spatial attention module is also embedded to enhance the deep feature representations of each path, thereby enhancing the deep feature representation of a path. For example, first calculate the channel attention weights. as follows: ; in, Global average pooling is used to compress the feature sequence of each channel into a single value to capture the overall response intensity of that channel. This is global max pooling, used to extract the peak response features of each channel; It is a multilayer perceptron; The sigmoid function is used. Then, the spatial attention weights are obtained. as follows: ; in, AvgPool(U) This represents the feature map, i.e., the deep feature representation. U Perform global average pooling along the channel dimension to generate a pool with shape... R 1×L A spatial feature map, where each element reflects the average response intensity of that location across all channels, is used to locate regions with high global activation. MaxPool(U) This indicates that global max pooling is performed on the feature map, i.e., the deep feature representation U, along the channel dimension to generate a shape of... R 1×L The spatial feature map, where each element reflects the maximum response intensity across all channels at that location, is used to highlight local salient features; and [ AvgPool(U) ; MaxPool(U) This process represents a splicing operation along the channel dimension, combining the above... AvgPool(U) The obtained spatial feature map and MaxPool(U) The obtained spatial feature maps are stitched together as follows R 2×L This is a one-dimensional convolution with a kernel size of 7. The final result is an attention-weighted deep feature representation. The relevant formulas are as follows: ; The deep feature representations, namely the high-frequency deep feature representation, the low-frequency deep feature representation, and the full-band deep feature representation, are weighted and concatenated in the fusion layer. Then, a one-dimensional convolutional layer is used to output the wavelet coefficients and the reconstructed running state signal. .

[0053] The reconstructed operating state wavelet coefficients and reconstructed operating state signals are obtained in the above embodiments. Then, from the reconstructed running status signal Extract the active segment of the reconstructed running state and reconfiguration of operating state noise segment and from the running status signal Extracting the original operating noise segment Based on the reconstructed running status active segment Obtain effective signal power And based on the reconstructed operating state noise segment and the original operating noise segment Obtaining noise distortion power Then, the operating status signal-to-noise ratio is generated based on the effective signal power and noise distortion power. The relevant formulas are as follows: ; in, A function for calculating the variance of the signal; reconstructing the active segment of the running state. To reconstruct the running status signal The extracted active segments that are determined to contain valid device status information; reconstructed operating status noise segments. To reconstruct the running status signal Extract the noise segment determined to contain only background noise; original operating state noise segment. Operating status signal The extracted noise segments are determined to contain only background noise; and the aforementioned noise segments are usually selected during periods of steady-state equipment operation or periods known to be free of abnormalities.

[0054] Based on the operating status signal and reconstruct running status signals The accumulated distance of the signal is obtained under the preset Dynamic Time Warping (DTW) algorithm; based on the running status signal. Obtain the original signal length and the standard deviation of the original signal Then, based on the original signal length and the standard deviation of the original signal Acquire the cumulative fluctuation of the original signal Then, based on the signal accumulation distance and the original signal accumulation fluctuation, a state waveform similarity is generated. The relevant formulas are as follows: ; Among them, state waveform similarity The closer to 1, the higher the morphological fidelity. State energy concentration is generated based on the wavelet coefficients of the running state and the reconstructed running state. The relevant formulas are as follows: ; in, C i This indicates that after sorting the wavelet coefficients in descending order of amplitude, the i-th coefficient in the numerator is... K Usually, the top 10% of the total number of coefficients are taken, that is... K=0.1N ,and N This represents the total number of wavelet coefficients. C j Represents the first wavelet coefficient among all wavelet coefficients. j There are several coefficients. And the numerator... For the front K The energy of the wavelet coefficients with the largest amplitude; denominator This represents the total energy of all wavelet coefficients. It can be seen that the state energy concentration... EC The higher the value, the better. EC The closer the value is to 1, the more concentrated the signal energy is in a few large coefficients, meaning the signal characteristics are prominent and there is less random noise; conversely, the closer the value is to 1, the more concentrated the signal energy is in a few large coefficients, meaning the signal has prominent characteristics and less random noise. EC The lower the value, the more energy is dispersed among a large number of small coefficients, resulting in more residual broadband noise or artifacts.

[0055] State energy concentration A higher value indicates a smoother signal or more prominent features, with less random noise. This is based on the operating status signal. and reconstruct running status signals Obtain the original low-frequency trend items respectively and reconstructing low-frequency trend items The aforementioned low-frequency trend terms can all be extracted using a low-pass filter, such as a Butterworth filter with an extremely low cutoff frequency, or by the moving average method.

[0056] Based on the above original low-frequency trend items and reconstructing low-frequency trend items Obtain the original low-frequency trend standard deviation and reconstructing the standard deviation of low-frequency trends And based on the original low-frequency trend standard deviation and reconstructing the standard deviation of low-frequency trends Obtain the degree of trend fluctuation Based on the original low-frequency trend term and reconstructing low-frequency trend items Obtain low-frequency trend correlation ( , Furthermore, based on low-frequency trend correlation... ( , and degree of trend fluctuation Generate state trend consistency The relevant formulas are as follows: ; Subsequently, based on the signal-to-noise ratio of the operating status... State waveform similarity State energy concentration Consistency with state trend Constructing a four-dimensional quality feature vector as follows: ; Any degradation in any aspect, such as oversmoothing leading to shape distortion or residual noise causing energy dispersion, will be reflected in the constructed quality feature vector, thus avoiding misjudgments that might occur when using a single metric. The quality feature vector... The input is classified into a pre-trained GBDT (Gradient Boosting Decision Tree) model, and the output is "Excellent", "Good", "Medium", or "Poor" as the corresponding signal quality level. The relevant formulas for the above GBDT model are as follows: ; in, This is the m-th tree in GBDT; The corresponding weights for GBDT; The total number of GBDT trees; the GBDT training objective is to minimize a loss function such as multi-class logarithmic loss. High-quality data of reconstructed operating status signals, whose signal quality level is classified as "excellent" or "good", are used as valid operating status signals, and fault operation and maintenance of the target converter transformer are performed based on the valid operating status signals.

[0057] In one embodiment, data from reconstructed operating status signals whose signal quality level is classified as "medium" or "poor" will trigger an alarm or suggest manual review.

[0058] The advantages of using the GBDT model for classification in the above embodiments are as follows: GBDT, by integrating multiple decision trees, can automatically learn the complex, non-linear interactions between the four evaluation indicators and their mapping to the final quality level. Compared to manually setting thresholds or using linear models such as logistic regression, it is more adaptable to the complex quality distribution boundaries in actual industrial data, and GBDT possesses better non-linear decision-making capabilities. GBDT is insensitive to the scale of input features and has good tolerance for outliers in features that may be caused by accidental interference in quality assessment. This ensures the stability and reliability of the grading system, thus GBDT is robust to features. The decision-making process of GBDT can be traced back to a series of indicator-based "if-then" rules, which makes the quality grading results no longer a black box. Operation and maintenance personnel can understand why the model classifies a certain signal as "good" rather than "excellent," improving the credibility and acceptability of the system in practical applications. Therefore, GBDT has interpretability and trustworthiness.

[0059] After performing fault maintenance on the target converter transformer, historical operating status signals, historical signal quality levels, and historical network parameter sets are acquired; based on the historical operating status signals, historical pure noise segments are extracted. Based on historical pure noise segments The optimized tradeoff factor is obtained by fitting a generalized Gaussian distribution and estimating the maximum likelihood. and optimized safety factor The probability density function corresponding to the generalized Gaussian distribution fitting is: ; in, Fit the probability density function to the generalized Gaussian distribution; The probability density function is preset with a mean, specifically a location parameter, which represents the mean level of the noise distribution and is usually 0; n A standard deviation is preset for the probability density function, specifically the amplitude random variable of the historical pure noise segment, i.e., the noise sample value. set up This is a scale parameter that reflects the degree of noise amplitude expansion. The larger the value, the stronger the noise energy. set up Let be the shape parameter, describing the attenuation characteristics of the noise distribution tail. It degenerates into a Gaussian distribution. When the value is 1, it follows a Laplace distribution. Maximum likelihood estimation is used to obtain the historical pure noise segment. The parameters were obtained by fitting. and And then based on The preset trade-off factor is dynamically adjusted in relation to a preset noise energy threshold. For example, the stronger the noise energy, the larger the trade-off factor is to strengthen the emphasis of wavelet basis matching on noise suppression. The preset safety factor is dynamically adjusted in relation to the preset distribution pattern. ， For example, the heavier the noise distribution tail, the greater the safety factor should be to reserve a wider frequency band protection margin, thereby obtaining an optimized trade-off factor and an optimized safety factor.

[0060] Then, a hyperparameter space including learning rate, batch size, path depth, etc., is defined. An observation dataset is constructed based on the historical signal quality level and the historical network parameter set. Based on the observed dataset, Bayesian optimization is performed under a preset maximum acquisition algorithm to obtain an optimized network parameter set. ; ; in, This is a preset maximum acquisition function. Based on the optimized network parameter set. Optimization is performed by configuring a preset multi-path neural network, specifically by configuring and training the network and then evaluating its performance. Optimize the current network parameter set Add the observation dataset, repeat the process of configuring and training the network iteratively until the preset number of evaluations or performance convergence is reached, output the optimal hyperparameter combination, deploy the preset multipath neural network, and obtain the optimized multipath neural network; use the optimized trade-off factor, the optimized safety factor, and the optimized multipath neural network as the trade-off factor, safety factor, and multipath neural network for the next execution of the converter transformer fault operation and maintenance method.

[0061] Please see Figure 2The present invention also provides a converter transformer fault operation and maintenance system for implementing any of the converter transformer fault operation and maintenance methods described above, comprising: an operation status signal acquisition module for acquiring the operation status signal of the target converter transformer; a signal and wavelet coefficient reconstruction module for obtaining operation status wavelet coefficients by frequency band decomposition of the operation status signal under a preset wavelet decomposition algorithm, and obtaining reconstructed operation status wavelet coefficients and a reconstructed operation status signal based on the operation status wavelet coefficients under a preset signal optimization algorithm; an active segment noise segment extraction module for extracting reconstructed operation status active segments and reconstructed operation status noise segments from the reconstructed operation status signal, and extracting original operation status noise segments from the operation status signal; a status signal-to-noise ratio generation module for generating an operation status signal-to-noise ratio based on the reconstructed operation status active segments, the reconstructed operation status noise segments, and the original operation status noise segments; and an accumulated fluctuation and distance generation module for generating a cumulative fluctuation and distance based on the operation status signal and the reconstructed operation status signal at a preset dynamic normalization distance. The algorithm acquires the cumulative distance of the signal and obtains the cumulative fluctuation of the original signal based on the operating state signal; a waveform similarity generation module is used to generate state waveform similarity based on the cumulative distance of the signal and the cumulative fluctuation of the original signal; an energy concentration generation module is used to generate state energy concentration based on the wavelet coefficients of the operating state and the reconstructed operating state; a trend consistency generation module is used to obtain the original low-frequency trend term and the reconstructed low-frequency trend term based on the operating state signal and the reconstructed operating state signal, and generate state trend consistency based on the original low-frequency trend term and the reconstructed low-frequency trend term; a signal quality level acquisition module is used to obtain the signal quality level based on the operating state signal-to-noise ratio, the state waveform similarity, the state energy concentration, and the state trend consistency; a fault operation and maintenance module is used to filter the reconstructed operating state signal according to the signal quality level to obtain a valid operating state signal, and perform fault operation and maintenance on the target converter transformer based on the valid operating state signal.

[0062] The above embodiments obtain the reconstructed operating state wavelet coefficients and the reconstructed operating state signal from the operating state signal contaminated by strong interference noise. Based on the reconstructed operating state signal and the original operating state signal, corresponding active and noise segments are extracted to generate the operating state signal-to-noise ratio (SNR). This SNR can reduce the possibility of misjudging or missing subsequent transformer fault symptoms due to strong background noise interference. A higher SNR indicates less noise contamination of the reconstructed signal. However, the operating state SNR ignores the assessment of the similarity of the reconstructed signal's morphology. Therefore, this embodiment generates a state waveform similarity based on the signal accumulation distance and the original signal accumulation fluctuation to effectively verify whether the reconstruction process distorts or loses key waveform features related to the converter transformer's state changes, such as transient impacts and oscillations. If the state waveform similarity is low, it indicates that the reconstructed signal has lost the physical morphological details required for fault identification. This embodiment also generates a state waveform similarity based on the operating state wavelet coefficients and the reconstructed operating state wavelet coefficients. The energy concentration of the state is used to identify broadband noise or artifacts remaining in the reconstructed signal. The higher the energy concentration, the more effectively the irregular interference energy scattered across various frequency bands has been removed, and the more focused the signal characteristics are. In this embodiment, the low-frequency trend terms corresponding to the operating state signal and the reconstructed operating state signal are extracted to generate the state trend consistency. Since the state trend consistency focuses on the low-frequency slow-change components of the signal, it can reflect the consistency of the trend information of the transformer state slow-change process before and after the signal reconstruction, which complements the signal-to-noise ratio, similarity, and energy concentration that focus on instantaneous or local features. The signal quality level obtained by this embodiment based on the above-mentioned operating state signal-to-noise ratio, state waveform similarity, state energy concentration, and state trend consistency has high reliability. Screening the reconstructed signal based on this level can improve the reliability of the obtained effective operating state signal. In turn, using the effective operating state signal for fault operation and maintenance can significantly improve the reliability of converter transformer fault operation and maintenance.

[0063] It is understood that the above system item embodiments correspond to the method item embodiments of the present invention, and can implement the converter transformer fault operation and maintenance method provided by any of the above method item embodiments of the present invention.

[0064] It should be noted that the system embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0065] Based on the above embodiments of the converter transformer fault operation and maintenance method, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the converter transformer fault operation and maintenance method of any embodiment of the present invention.

[0066] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0067] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0068] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0069] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for fault operation and maintenance of a converter transformer, characterized in that, Includes the following steps: Acquire the operating status signal of the target converter transformer; Based on the operating state signal, frequency band decomposition under a preset wavelet decomposition algorithm is used to obtain operating state wavelet coefficients, and based on the operating state wavelet coefficients, reconstructed operating state wavelet coefficients and reconstructed operating state signal are obtained under a preset signal optimization algorithm. The reconstructed running state active segment and the reconstructed running state noise segment are extracted from the reconstructed running state signal, and the original running state noise segment is extracted from the running state signal; The signal-to-noise ratio of the operating state is generated based on the reconstructed operating state activity segment, the reconstructed operating state noise segment, and the original operating state noise segment; Based on the operating status signal and the reconstructed operating status signal, the cumulative signal distance is obtained under a preset dynamic normalization distance algorithm, and the cumulative fluctuation of the original signal is obtained based on the operating status signal; The similarity of the generated state waveform is based on the signal accumulation distance and the original signal accumulation fluctuation; The state energy concentration is generated based on the wavelet coefficients of the operating state and the reconstructed operating state. Based on the operating status signal and the reconstructed operating status signal, the original low-frequency trend item and the reconstructed low-frequency trend item are obtained, and the state trend consistency degree is generated based on the original low-frequency trend item and the reconstructed low-frequency trend item. The signal quality level is obtained based on the operating state signal-to-noise ratio, the state waveform similarity, the state energy concentration, and the state trend consistency. The reconstructed operating status signals are filtered according to the signal quality level to obtain valid operating status signals, and the target converter transformer is subjected to fault maintenance based on the valid operating status signals.

2. The method for fault operation and maintenance of a converter transformer as described in claim 1, characterized in that, The step of generating the operating state signal-to-noise ratio based on the reconstructed operating state activity segment, the reconstructed operating state noise segment, and the original operating state noise segment includes: The effective signal power is obtained based on the reconstructed operating state active segment, and the noise distortion power is obtained based on the reconstructed operating state noise segment and the original operating state noise segment; The operating state signal-to-noise ratio is generated based on the effective signal power and noise distortion power.

3. The method for fault operation and maintenance of a converter transformer as described in claim 2, characterized in that, The step of obtaining the signal cumulative distance based on the operating status signal and the reconstructed operating status signal under a preset dynamic warping distance algorithm, and obtaining the original signal cumulative fluctuation based on the operating status signal, includes: Based on the operating status signal and the reconstructed operating status signal, the cumulative signal distance is obtained under a preset dynamic warping distance algorithm; The original signal length and original signal standard deviation are obtained based on the operating status signal; The cumulative fluctuation of the original signal is obtained based on the original signal length and the original signal standard deviation.

4. The method for fault operation and maintenance of a converter transformer as described in claim 3, characterized in that, The step of obtaining the original low-frequency trend term and the reconstructed low-frequency trend term based on the operating state signal and the reconstructed operating state signal, and generating a state trend consistency degree based on the original low-frequency trend term and the reconstructed low-frequency trend term, includes: Based on the operating status signal and the reconstructed operating status signal, the original low-frequency trend term and the reconstructed low-frequency trend term are obtained; Based on the original low-frequency trend term and the reconstructed low-frequency trend term, the standard deviation of the original low-frequency trend and the standard deviation of the reconstructed low-frequency trend are obtained, and the degree of trend fluctuation is obtained based on the standard deviation of the original low-frequency trend and the standard deviation of the reconstructed low-frequency trend. The low-frequency trend correlation is obtained based on the original low-frequency trend term and the reconstructed low-frequency trend term; A state trend consistency degree is generated based on the low-frequency trend correlation and the trend fluctuation degree.

5. A method for fault operation and maintenance of a converter transformer as described in claim 1, characterized in that, The step of obtaining operating state wavelet coefficients through frequency band decomposition using a preset wavelet decomposition algorithm based on the operating state signal, and obtaining reconstructed operating state wavelet coefficients and a reconstructed operating state signal based on the operating state wavelet coefficients using a preset signal optimization algorithm, includes: The key frequency band distribution features, signal stationarity features, and singularity distribution features are extracted from the time-frequency analysis of the operating status signal; Based on the key frequency band distribution characteristics, the signal stationarity characteristics, and the singularity distribution characteristics, the optimal operating state wavelet basis is obtained by screening under a preset wavelet function library and a preset trade-off factor. The power spectrum of the state signal is obtained by performing a Fourier transform based on the operating state signal, and the lowest frequency of the effective signal is obtained based on the power spectrum of the state signal. The optimal number of decomposition layers is generated based on the lowest frequency of the effective signal under a preset sampling frequency and a preset safety factor. Based on the optimal decomposition level and the optimal operating state wavelet basis, the operating state signal is decomposed into wavelet coefficients and approximation coefficients to obtain detail coefficients and approximation coefficients, and the detail coefficients and approximation coefficients are integrated to obtain the operating state wavelet coefficients. Based on the wavelet coefficients of the operating state, the reconstructed operating state wavelet coefficients and the reconstructed operating state signal are obtained under a preset signal optimization algorithm.

6. A method for fault operation and maintenance of a converter transformer as described in claim 5, characterized in that, The optimal operating state wavelet basis is obtained by screening based on the key frequency band distribution characteristics, the signal stationarity characteristics, and the singular point distribution characteristics under a preset wavelet function library and a preset tradeoff factor, including: Based on the key frequency band distribution characteristics, the signal stationarity characteristics, and the singularity distribution characteristics, a candidate set of wavelet basis functions for the running state is selected from a preset wavelet function library; For each wavelet basis in the candidate set of wavelet basis for the operating state, a signal matching index is generated for each wavelet basis under the operating state signal and a preset trade-off factor. The optimal operating state wavelet basis is obtained by screening the candidate set of the operating state wavelet basis based on the signal matching degree index.

7. A method for fault operation and maintenance of a converter transformer as described in claim 6, characterized in that, The process of obtaining the reconstructed operating state wavelet coefficients and the reconstructed operating state signal based on the operating state wavelet coefficients under a preset signal optimization algorithm includes: The detailed coefficients of the wavelet coefficients in the running state are convolved and mapped under the high-frequency path of a preset multi-path neural network to obtain a high-frequency deep feature representation. The approximate coefficients of the wavelet coefficients of the running state are processed by convolution mapping under the low-frequency path of the preset multi-path neural network to obtain the low-frequency deep feature representation. The operating status signal is processed by convolutional mapping under the full-band path of a preset multi-path neural network to obtain a full-band deep feature representation; The wavelet coefficients and reconstructed operating state signal are obtained by splicing together the high-frequency deep feature representation, the low-frequency deep feature representation, and the full-band deep feature representation.

8. A method for fault operation and maintenance of a converter transformer as described in claim 7, characterized in that, After filtering the reconstructed operating status signals according to signal quality levels to obtain valid operating status signals, and performing fault maintenance on the target converter transformer based on the valid operating status signals, the process further includes: Acquire historical operating status signals, historical signal quality levels, and historical network parameter sets; Based on the historical operating status signal, a historical pure noise segment is extracted, and based on the historical pure noise segment, a generalized Gaussian distribution fitting and maximum likelihood estimation are performed to obtain the optimization trade-off factor and the optimization safety factor. An observation dataset is constructed based on the historical signal quality level and the historical network parameter set, and an optimized network parameter set is obtained by performing Bayesian optimization on the observation dataset under a preset maximum acquisition algorithm. The preset multipath neural network is optimized based on the optimized network parameter set to obtain the optimized multipath neural network; The optimization trade-off factor, the optimization safety factor, and the optimization multipath neural network are used as the trade-off factor, safety factor, and multipath neural network for the next execution of the converter transformer fault operation and maintenance method.

9. A converter transformer fault operation and maintenance system, characterized in that, A method for implementing a converter transformer fault operation and maintenance as described in any one of claims 1 to 8 includes: The operation status signal acquisition module is used to acquire the operation status signals of the target converter transformer; The signal and wavelet coefficient reconstruction module is used to obtain the operating state wavelet coefficients by frequency band decomposition under a preset wavelet decomposition algorithm based on the operating state signal, and to obtain the reconstructed operating state wavelet coefficients and the reconstructed operating state signal based on the operating state wavelet coefficients under a preset signal optimization algorithm. The active segment noise segment extraction module is used to extract the reconstructed operating state active segment and the reconstructed operating state noise segment from the reconstructed operating state signal, and to extract the original operating state noise segment from the operating state signal; The state signal-to-noise ratio generation module is used to generate the operating state signal-to-noise ratio based on the reconstructed operating state activity segment, the reconstructed operating state noise segment, and the original operating state noise segment; The cumulative fluctuation and distance generation module is used to obtain the signal cumulative distance based on the operating status signal and the reconstructed operating status signal under a preset dynamic normalization distance algorithm, and to obtain the original signal cumulative fluctuation based on the operating status signal; A waveform similarity generation module is used to generate state waveform similarity based on the signal cumulative distance and the original signal cumulative fluctuation. An energy concentration generation module is used to generate state energy concentration based on the wavelet coefficients of the operating state and the reconstructed operating state wavelet coefficients. The trend consistency generation module is used to obtain the original low-frequency trend item and the reconstructed low-frequency trend item based on the running state signal and the reconstructed running state signal, and to generate the state trend consistency degree based on the original low-frequency trend item and the reconstructed low-frequency trend item. The signal quality level acquisition module is used to acquire the signal quality level based on the operating state signal-to-noise ratio, the state waveform similarity, the state energy concentration, and the state trend consistency. The fault operation and maintenance module is used to filter the reconstructed operating status signal according to the signal quality level to obtain a valid operating status signal, and to perform fault operation and maintenance on the target converter transformer based on the valid operating status signal.

10. A terminal device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the converter transformer fault operation and maintenance method as described in any one of claims 1 to 8.

Images