Transformer core looseness diagnosis method based on voiceprint spectrum and data enhancement

Abstract

The application relates to a transformer core loosening diagnosis method based on a voiceprint spectrum and data enhancement, relates to the technical field of transformer fault diagnosis, and comprises the following steps: collecting core voiceprint signals and winding electrical parameter sets of a target transformer; intercepting a voiceprint signal segment on the core voiceprint signals and performing data enhancement processing to obtain an enhanced core voiceprint spectrum and identify and acquire core operation characteristics; identifying core loosening abnormalities according to the core operation characteristics to obtain core loosening abnormal characteristics; verifying the core loosening abnormal characteristics based on the winding electrical parameter sets, and issuing a core loosening fault signal according to the verification result. The application solves the problems in traditional transformer core loosening diagnosis, such as single monitoring mode, lack of effective verification of diagnosis results, frequent false alarms caused by interference, insufficient diagnosis accuracy and reliability, and difficulty in meeting the precise transformer fault monitoring demand of a power system.

Description

Technical Field

[0001] This application relates to the field of transformer fault diagnosis, and in particular to a method for diagnosing transformer core loosening based on acoustic signature mapping and data enhancement. Background Technology

[0002] With the increasing demands for power grid security and power supply reliability, accurate diagnosis of transformer core loosening faults has become a key technical requirement to ensure stable equipment operation.

[0003] Currently, traditional methods for diagnosing transformer core loosening have limitations such as a single monitoring method, a lack of effective verification of diagnostic results, susceptibility to external interference leading to frequent false alarms, and insufficient diagnostic accuracy and reliability, failing to meet the actual needs of power systems for efficient and accurate monitoring of transformer faults. Summary of the Invention

[0004] This application provides a method for diagnosing transformer core loosening based on acoustic signature and data enhancement, which improves the current situation of traditional transformer core loosening diagnosis, which has the problems of single monitoring method, lack of effective verification of diagnostic results, susceptibility to interference leading to frequent false alarms, and insufficient diagnostic accuracy and reliability.

[0005] The embodiments of this application disclose the following technical solutions:

[0006] This application provides a method for diagnosing transformer core loosening based on acoustic signature mapping and data enhancement. The method includes:

[0007] Acoustic fingerprint signals of the transformer core of the target transformer are collected to obtain the core acoustic fingerprint signal, and the electrical parameters of the transformer winding of the target transformer are collected simultaneously to obtain the winding electrical parameter set;

[0008] A segment of acoustic pattern signal is extracted from the acoustic pattern signal of the iron core, and the acoustic pattern signal segment is subjected to data enhancement processing to obtain an enhanced iron core acoustic pattern spectrum. The iron core operation characteristics are then identified based on the enhanced iron core acoustic pattern spectrum to obtain the iron core operation characteristics.

[0009] Based on the described core operating characteristics, core loosening anomalies are identified to obtain core loosening anomaly characteristics;

[0010] The abnormal characteristics of core loosening are verified based on the set of winding electrical parameters, and a core loosening fault signal is issued according to the verification results.

[0011] Optionally, the acoustic signature signal of the transformer core of the target transformer is acquired to obtain the core acoustic signature signal, including:

[0012] Acoustic sensors are arranged at multiple locations in the transformer core to form an acoustic sensor array;

[0013] According to the preset sampling frequency, the acoustic signature signal of the transformer core is collected by the acoustic signature sensor array to obtain the core acoustic signature signal.

[0014] Optionally, the electrical parameters of the transformer windings of the target transformer are collected synchronously to obtain a set of winding electrical parameters, including:

[0015] Obtain preset winding electrical parameters, and configure a winding electrical parameter monitoring unit for the transformer winding based on the preset winding electrical parameters;

[0016] According to the preset sampling frequency, the electrical parameters of the transformer winding are collected by the winding electrical parameter monitoring unit to obtain the winding electrical parameter set.

[0017] Optionally, the preset winding electrical parameters include at least one of excitation current harmonics, three-phase current imbalance, and winding vibration signal.

[0018] Optionally, a segment of the acoustic signature signal is extracted from the acoustic signature signal of the iron core, and data enhancement processing is performed on the acoustic signature signal segment to obtain an enhanced acoustic signature spectrum of the iron core, including:

[0019] A preset judgment window is set, and a segment of the acoustic pattern signal is extracted from the acoustic pattern signal of the iron core based on the preset judgment window;

[0020] Based on the instantaneous signal-to-noise ratio and frequency change rate of the voiceprint signal segment, the window length parameter of the short-time Fourier transform is dynamically adjusted to perform time-frequency analysis on the voiceprint signal segment and obtain time-frequency domain data.

[0021] The time-frequency domain data is filtered by a trainable Mel filter bank and converted into a two-dimensional acoustic signature.

[0022] The two-dimensional acoustic signature spectrum is enhanced to obtain an enhanced iron core acoustic signature spectrum. The enhancement process includes weighted processing based on spectral entropy and random time-frequency masking processing.

[0023] Optionally, the core operating characteristics are identified based on the enhanced core acoustic signature, and the core operating characteristics are obtained, including:

[0024] The running feature recognizer is retrieved, which is constructed based on the sample enhanced iron core acoustic pattern spectrum set and the sample iron core running feature set;

[0025] The enhanced core acoustic signature is input into the operating feature recognizer to identify the core operating features and output the core operating features.

[0026] Optionally, based on the core operating characteristics, core loosening anomaly identification is performed to obtain core loosening anomaly characteristics, including:

[0027] Collect historical normal operation feature sets of transformer cores under normal conditions, and establish a normal operation feature template of the core based on the historical normal operation feature sets of the cores;

[0028] The operating characteristics of the iron core are compared and analyzed with the normal operating characteristics template of the iron core to calculate the operating characteristic deviation;

[0029] When the deviation of the operating characteristics exceeds the preset abnormal threshold, the operating characteristics of the iron core are regarded as abnormal characteristics of iron core loosening.

[0030] Optionally, the abnormal characteristics of core loosening are verified based on the winding electrical parameter set, and a core loosening fault signal is issued according to the verification results, including:

[0031] Obtain the abnormal timestamp interval of the core loosening abnormality feature, and extract the corresponding electrical parameters from the winding electrical parameter set based on the abnormal timestamp interval to obtain the electrical parameter set to be verified;

[0032] Feature extraction is performed on the set of electrical parameters to be verified to obtain the electrical characteristics of the winding to be verified;

[0033] The winding response analyzer is invoked, and the abnormal characteristics of the loose core are processed based on the winding response analyzer to obtain the winding response electrical characteristics.

[0034] Calculate the similarity between the electrical characteristics of the winding to be verified and the electrical characteristics of the winding response, and obtain the electrical characteristic similarity as the verification result;

[0035] When the electrical feature similarity is greater than or equal to a preset similarity threshold, a core loosening fault signal is issued.

[0036] Optionally, the construction steps of the winding response analyzer include:

[0037] Obtain the transformer model of the target transformer, and retrieve multiple historical core loosening anomaly records of transformers of the same model based on the transformer model. Each core loosening anomaly record includes historical core loosening anomaly characteristics and historical winding electrical characteristics.

[0038] A sample core loosening anomaly feature set is constructed based on the historical core loosening anomaly features in multiple historical core loosening anomaly records, and a sample winding response electrical feature set is constructed based on the historical winding electrical features in multiple historical core loosening anomaly records;

[0039] Using the sample core loosening anomaly feature set as input features and the sample winding response electrical feature set as output target, the winding response analyzer is trained and generated.

[0040] Optionally, when the electrical feature similarity is less than a preset similarity threshold, an abnormal core state signal is issued.

[0041] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0042] This application proposes a method for diagnosing transformer core loosening based on acoustic signature mapping and data enhancement. By synchronously collecting core acoustic signatures and winding electrical data in stages, generating enhanced core acoustic signature mapping, constructing an operational feature identifier to identify anomalies, calling a winding response analyzer to verify anomalies, and outputting signals in stages, the method achieves accurate diagnosis of transformer core loosening faults. First, the acoustic signature signal of the target transformer core and the electrical parameters of the windings are collected simultaneously, ensuring that both data use a unified preset sampling frequency. Then, the core acoustic signature signal is segmented according to a preset judgment window. The short-time Fourier transform window length is dynamically adjusted according to the instantaneous signal-to-noise ratio and frequency change rate of the signal segment to achieve adaptive time-frequency analysis. After being converted into a two-dimensional acoustic signature spectrum by a trainable Mel filter bank, data enhancement is completed through weighted processing based on spectral entropy and random time-frequency occlusion processing to obtain an enhanced core acoustic signature spectrum. Subsequently, the enhanced core acoustic signature spectrum is input into a pre-constructed operating feature recognizer to extract the core operating features and compare it with the normal operating feature template constructed based on historical normal data. If the operating feature deviation exceeds a preset threshold, it is identified as a core loosening anomaly, and the abnormal features are extracted. Finally, according to the timestamp interval of the core loosening anomaly, the corresponding time period parameters are extracted from the winding electrical parameter set, and the electrical features of the winding to be verified are extracted. The winding response analyzer constructed based on the historical core loosening records of the same type of transformer is called to obtain the expected response features. The similarity between the two types of features is calculated, and a core loosening fault signal or core abnormality signal is issued according to a preset threshold.

[0043] The technical solution proposed in this application solves the problems of false alarms caused by traditional single monitoring methods, difficulty in extracting key features from complex core acoustic signals, and lack of multi-source data verification support for diagnostic results. It avoids misjudgment and missed judgment of faults caused by interference from a single signal or lack of cross-verification, effectively improves the identification accuracy and reliability of transformer core loosening fault diagnosis, and provides technical support for transformer operation and maintenance decisions in power systems. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1A schematic flowchart of the transformer core loosening diagnosis method based on acoustic signature and data enhancement provided in the embodiments of this application;

[0046] Figure 2 This is a schematic diagram illustrating the process of obtaining the acoustic signature of the reinforced iron core, as provided in an embodiment of this application. Detailed Implementation

[0047] This application provides a method for diagnosing transformer core loosening based on acoustic signature and data enhancement, which addresses the technical problems of existing traditional transformer core loosening diagnosis and monitoring methods, such as being simplistic, lacking effective verification of diagnostic results, being susceptible to environmental interference leading to frequent false alarms, and having insufficient diagnostic accuracy and reliability.

[0048] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and 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.

[0049] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0050] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0051] Examples, as shown in the appendix Figure 1 As shown, this application provides a method for diagnosing transformer core loosening based on acoustic signature mapping and data enhancement. The method includes the following steps:

[0052] S110: Collect the acoustic signature signal of the transformer core of the target transformer to obtain the core acoustic signature signal, and simultaneously collect the electrical parameters of the transformer winding of the target transformer to obtain the winding electrical parameter set;

[0053] In this embodiment of the application, in the scenario of online monitoring and diagnosis of transformer core loosening faults, in order to simultaneously acquire the characteristic signals that reflect the source of the core fault and the winding association signals that can be used for collaborative verification, it is necessary to simultaneously collect the acoustic fingerprint signal of the transformer core and the electrical parameters of the windings to build a diagnostic basis for multi-source data fusion.

[0054] Specifically, for the acquisition of acoustic signature signals of the iron core, acoustic signature sensors need to be deployed at multiple locations on the transformer iron core. By deploying acoustic signature sensor arrays at multiple locations to cover key areas of the iron core, it is possible to ensure that acoustic signature changes caused by loosening in different parts of the iron core can be fully captured, avoiding signal omissions caused by single-point acquisition.

[0055] Subsequently, the acoustic signature signal of the transformer core is continuously collected through the acoustic signature sensor array according to the preset sampling frequency, and finally the acoustic signature signal of the core that can reflect the operating status of the core in real time is obtained.

[0056] Meanwhile, for the acquisition of winding electrical parameters, a dedicated winding electrical parameter monitoring unit needs to be configured for the transformer winding based on the preset winding electrical parameters to ensure that the acquisition range and accuracy of the monitoring unit match the monitoring requirements of the preset parameters.

[0057] Furthermore, following the same preset sampling frequency as the acoustic signature signal acquisition, the electrical parameters of the transformer winding are synchronously acquired through the winding electrical parameter monitoring unit to ensure that the two types of data maintain consistency in the time dimension, and finally obtain the winding electrical parameter set that matches the time of the core acoustic signature signal.

[0058] This step, by simultaneously acquiring the core acoustic signature signal and winding electrical parameters, not only obtains the direct characteristic signal of the fault source but also obtains the associated electrical signal that can be used for subsequent verification. This effectively makes up for the shortcomings of traditional single monitoring methods in terms of one-sided information and lays a data foundation for improving diagnostic accuracy through data augmentation, feature recognition, and collaborative verification.

[0059] Step S110 in the method provided in this application embodiment includes:

[0060] Acoustic sensors are arranged at multiple locations in the transformer core to form an acoustic sensor array;

[0061] According to the preset sampling frequency, the acoustic signature signal of the transformer core is collected by the acoustic signature sensor array to obtain the core acoustic signature signal.

[0062] Obtain preset winding electrical parameters, and configure a winding electrical parameter monitoring unit for the transformer winding based on the preset winding electrical parameters;

[0063] According to the preset sampling frequency, the electrical parameters of the transformer winding are collected by the winding electrical parameter monitoring unit to obtain the winding electrical parameter set.

[0064] In this embodiment of the application, in order to obtain the original signal that reflects the state of the transformer core and the associated parameters that can be used for collaborative verification, it is necessary to deploy the acoustic fingerprint sensor and configure the monitoring unit in a specific way, and collect the two types of data synchronously at a unified frequency to construct a multi-source and time-aligned basic dataset.

[0065] First, to collect acoustic fingerprint signals from the transformer core, acoustic fingerprint sensors need to be deployed at multiple locations on the transformer core to form an acoustic fingerprint sensor array.

[0066] Specifically, when deploying the array, it is necessary to consider the structural characteristics of the iron core and select key areas such as the iron core column and yoke that are prone to generating characteristic acoustic patterns due to loosening, so as to ensure that the array can fully capture the acoustic pattern changes of different parts of the iron core and avoid feature omissions caused by single-point acquisition.

[0067] Furthermore, after the acoustic sensor array is deployed, acoustic signals are collected from the transformer core at a preset sampling frequency to obtain the core acoustic signals.

[0068] The setting of the preset sampling frequency needs to be combined with the frequency range of the sound patterns that may be generated by the loose iron core. For example, for the low-frequency vibration sound patterns that are common in the loose iron core, the sampling frequency is set to a value that satisfies the Nyquist sampling theorem and can cover the low-frequency range. At the same time, the sampling frequency will be associated with the generation of timestamps. Each collected iron core sound pattern signal has a corresponding timestamp, which provides a basis for establishing a time correlation with the winding electrical parameters in the future.

[0069] Meanwhile, for the acquisition of transformer winding electrical parameters, it is first necessary to obtain preset winding electrical parameters in order to indirectly reflect changes in the core state.

[0070] In the method provided in this application embodiment, the preset winding electrical parameters include at least one of excitation current harmonics, three-phase current imbalance, and winding vibration signal.

[0071] Specifically, the preset winding electrical parameters may include at least one of the following: excitation current harmonics, three-phase current imbalance, and winding vibration signal. When selecting parameters, it is necessary to consider the possible impact of core loosening on the winding electrical characteristics to ensure that the selected parameters can effectively correlate with the abnormal state of the core.

[0072] For example, when the transformer core becomes loose, the symmetry of the core magnetic circuit will be disrupted, resulting in additional harmonic components in the excitation current. In this case, the excitation current harmonics are selected as the preset winding electrical parameters, and the loosening of the core can be indirectly reflected by monitoring the changes in harmonic content.

[0073] In addition, if the iron core is loose, it will cause uneven stress on the winding, resulting in an imbalance in the three-phase current distribution. The three-phase current imbalance can be used as a preset winding electrical parameter to capture the current fluctuations caused by the iron core abnormality.

[0074] In addition, the vibration caused by the loosening of the iron core will be transmitted to the winding, causing changes in the amplitude and frequency characteristics of the winding vibration signal. At this time, the winding vibration signal can be used as a preset winding electrical parameter directly related to the state of the iron core to help determine whether there is a loosening problem in the iron core.

[0075] Furthermore, based on the determined preset winding electrical parameters, a winding electrical parameter monitoring unit is configured for the transformer winding. During the configuration process, it is necessary to ensure that the measurement accuracy and range of the monitoring unit match the monitoring requirements of the selected electrical parameters. For example, for monitoring excitation current harmonics, a monitoring module that can accurately capture harmonic components is selected; for monitoring three-phase current imbalance, a unit that can simultaneously acquire three-phase currents and calculate the imbalance is configured to ensure that the monitoring unit can accurately obtain the preset winding electrical parameters.

[0076] Furthermore, after configuring the winding electrical parameter monitoring unit, the electrical parameters of the transformer winding are collected through the configured winding electrical parameter monitoring unit at the same preset sampling frequency as the core acoustic signal, thus obtaining the winding electrical parameter set.

[0077] Because a unified preset sampling frequency is used, the timestamp generated when the winding electrical parameters are collected is completely consistent with the timestamp of the core acoustic pattern signal. This ensures that the core acoustic pattern signal at each moment can correspond to the winding electrical parameters at the same time, thereby achieving accurate alignment of the two types of data in the time dimension. This effectively avoids the inability to establish the correlation between the core acoustic pattern and the winding electrical parameters in subsequent analysis due to time asynchrony.

[0078] Ultimately, the core acoustic signature signal and winding electrical parameter set obtained through the above steps effectively compensate for the limitations of traditional single monitoring data, laying a data foundation for subsequent data enhancement processing of acoustic signature signals, extraction of core operating characteristics, and verification of core loosening abnormalities based on winding electrical parameters.

[0079] S120: Extract a segment of acoustic pattern signal from the acoustic pattern signal of the iron core, perform data enhancement processing on the acoustic pattern signal segment to obtain an enhanced acoustic pattern spectrum of the iron core, and perform iron core operation feature identification based on the enhanced acoustic pattern spectrum of the iron core to obtain the iron core operation feature.

[0080] In this embodiment of the application, in order to effectively handle the complexity of long-term core acoustic signature signals and highlight the key features related to core loosening, the acoustic signature signals need to be processed through a process of segmented interception, adaptive time-frequency analysis, spectrum conversion enhancement, and feature recognition to obtain feature information that can accurately reflect the operating status of the core, providing a basis for subsequent identification of core loosening anomalies.

[0081] Specifically, a preset judgment window is first set, and based on this preset judgment window, the acoustic signal segment is extracted from the collected iron core acoustic signal to ensure that subsequent processing is more focused on the effective signal range.

[0082] Furthermore, after extracting the voiceprint signal segment, the window length parameter of the short-time Fourier transform is dynamically adjusted according to the instantaneous signal-to-noise ratio and frequency change rate of the voiceprint signal segment to perform time-frequency analysis on the voiceprint signal segment and obtain time-frequency domain data.

[0083] By dynamically adjusting the window length, a longer window length can be used in the range where the instantaneous signal-to-noise ratio is low and the frequency changes slowly to improve frequency resolution, while a shorter window length can be used in the range where the instantaneous signal-to-noise ratio is high and the frequency changes rapidly to improve time resolution, thereby capturing the time-frequency characteristics in the voiceprint signal more comprehensively.

[0084] Furthermore, a trainable Mel filter bank is used to filter the time-frequency domain data, converting the one-dimensional time-frequency domain data into a two-dimensional acoustic signature. The trainable Mel filter bank can optimize the filtering parameters according to the characteristics of the core acoustic signature signal, making the converted two-dimensional acoustic signature more closely match the expression requirements of the core's operating characteristics, and making key features easier to distinguish in the signature.

[0085] Furthermore, the two-dimensional acoustic signature is enhanced to obtain an enhanced core acoustic signature.

[0086] The enhancement processing includes weighted processing based on spectral entropy and random time-frequency occlusion processing. Weighted processing based on spectral entropy can increase the weight of frequency bands with high information entropy and containing key features in the spectrum, thereby highlighting the core features.

[0087] In addition, random time-frequency occlusion processing can enhance the robustness of the map to noise interference and reduce the interference impact during subsequent feature recognition.

[0088] Furthermore, an operational feature recognizer is retrieved. This recognizer is constructed based on the sample enhanced core acoustic signature spectrum set and the sample core operational feature set, and has the ability to extract operational features from the enhanced acoustic signature spectrum. The enhanced core acoustic signature spectrum is input into the operational feature recognizer for core operational feature recognition, and finally, the core operational features are output.

[0089] This step transforms the original acoustic signature signal into operational features that can be directly used for subsequent anomaly detection by segmenting, performing time-frequency analysis, spectrum conversion enhancement, and feature recognition on the iron core acoustic signature signal. This effectively solves the problems of unclear features and susceptibility to interference in the original acoustic signature signal.

[0090] As attached Figure 2 As shown, step S120 in the method provided in this application embodiment includes:

[0091] A preset judgment window is set, and a segment of the acoustic pattern signal is extracted from the acoustic pattern signal of the iron core based on the preset judgment window;

[0092] Based on the instantaneous signal-to-noise ratio and frequency change rate of the voiceprint signal segment, the window length parameter of the short-time Fourier transform is dynamically adjusted to perform time-frequency analysis on the voiceprint signal segment and obtain time-frequency domain data.

[0093] The time-frequency domain data is filtered by a trainable Mel filter bank and converted into a two-dimensional acoustic signature.

[0094] The two-dimensional acoustic signature spectrum is enhanced to obtain an enhanced iron core acoustic signature spectrum. The enhancement process includes weighted processing based on spectral entropy and random time-frequency masking processing.

[0095] The running feature recognizer is retrieved, which is constructed based on the sample enhanced iron core acoustic pattern spectrum set and the sample iron core running feature set;

[0096] The enhanced core acoustic signature is input into the operating feature recognizer to identify the core operating features and output the core operating features.

[0097] In this embodiment of the application, in order to convert the complex core acoustic signal into data that can accurately extract operating characteristics and avoid identification deviations caused by redundancy of the original signal or insufficient time-frequency resolution, the core acoustic signal needs to be processed through a process of segmentation, adaptive time-frequency analysis, spectrum conversion, enhancement processing and feature recognition to obtain feature information that can accurately reflect the operating status of the core, thus providing support for subsequent identification of core loosening anomalies.

[0098] Specifically, a preset judgment window is first set, and based on the preset judgment window, the acoustic pattern signal segment is extracted from the collected iron core acoustic pattern signal to decompose the continuous long-term iron core acoustic pattern signal into multiple independent signal segments containing complete acoustic feature cycles.

[0099] The preset judgment window length needs to be determined in combination with the characteristics of the iron core acoustic signal. For example, a sliding window of 1-5 seconds or a short time window of 25ms-50ms can be selected to ensure that each segment of acoustic signal can contain the complete acoustic feature cycle, while avoiding signal redundancy caused by an excessively long window.

[0100] Furthermore, after capturing the voiceprint signal segment, the window length parameter of the short-time Fourier transform is dynamically adjusted based on the instantaneous signal-to-noise ratio and frequency change rate of the signal segment to perform time-frequency analysis on the voiceprint signal segment and obtain time-frequency domain data. The instantaneous signal-to-noise ratio is determined by the ratio of signal power to noise power, and the frequency change rate is determined by calculating the gradient of the dominant frequency at adjacent time points.

[0101] Specifically, when the instantaneous signal-to-noise ratio is high and the frequency change rate is low, a larger window length is used to improve the frequency resolution and capture low-frequency stable features more clearly; when the instantaneous signal-to-noise ratio is low or the frequency change rate is high, a smaller window length is used to improve the time resolution and accurately capture high-frequency transient features. Through the adaptive adjustment of the above methods, it can be ensured that time-frequency analysis can fully cover different feature types of core acoustic signatures.

[0102] Furthermore, after completing the time-frequency analysis, the obtained time-frequency domain data is input into a trainable Mel filter bank for filtering and converted into a two-dimensional acoustic signature.

[0103] The trainable Mel filter bank contains multiple triangular filters whose center frequency and bandwidth parameters are not fixed values, but can be jointly optimized and adjusted with the deep learning model. That is, through the feedback during the model training process, it dynamically adapts to the characteristics of the target transformer core acoustic signal and the characteristic expression requirements of core loosening fault, so as to avoid the limitation that fixed parameter filters cannot match different transformer models or fault types.

[0104] Meanwhile, the Mel filter bank can convert the linear frequency distribution of the original time-frequency domain data into a Mel frequency distribution that better matches the human ear's perception of sound frequency, ultimately forming a two-dimensional acoustic graph with time as the horizontal axis and Mel frequency as the vertical axis, making key features such as low-frequency vibrations that may be caused by a loose iron core easier to distinguish in the graph.

[0105] Specifically, during the conversion process, a Short-Time Fourier Transform (STFT) must first be performed on each extracted voiceprint signal segment to obtain the corresponding spectral data. Then, the spectral data of each segment are sequentially input into a trainable Mel filter bank. Through frequency mapping and energy aggregation processing of the filter bank, the Mel features corresponding to each signal segment are obtained.

[0106] Finally, the Mel features of all signal segments are combined and spliced ​​in chronological order to form a complete two-dimensional acoustic signature map, which visually shows the feature distribution of the iron core acoustic signature signal at different times and frequencies, laying the foundation for subsequent capture of subtle feature changes related to iron core loosening.

[0107] Furthermore, the obtained two-dimensional acoustic signature is enhanced to obtain an enhanced core acoustic signature. This enhancement process includes two strategies.

[0108] Specifically, the first method is weighted processing based on spectral entropy. This involves calculating the spectral entropy value (a parameter reflecting spectral complexity) of each frequency band and assigning weights to different frequency components. Frequency bands with high spectral entropy values, potentially containing fault characteristics, receive increased weights to highlight weak fault-related features. The second method is random time-frequency occlusion processing. During training, time periods or frequency segments in the spectrum are randomly occluded, with an occlusion ratio of 5-20%. This forces the subsequent recognition model to learn the ability to extract features from partial information, thereby improving model generalization and reducing the impact of noise interference.

[0109] Furthermore, after obtaining the enhanced acoustic signature of the iron core, the operating feature recognizer is retrieved to accurately extract the operating features of the iron core from the enhanced acoustic signature.

[0110] The operational feature recognizer needs to be pre-constructed based on a sample enhanced core acoustic signature spectrum set and a sample core operational feature set. Through extensive sample training, the recognizer is trained to extract effective features from the enhanced core acoustic signature spectrum. The enhanced core acoustic signature spectrum is then input into the operational feature recognizer, which automatically performs feature extraction and analysis, ultimately outputting core operational features that reflect the current state of the core.

[0111] For example, when a certain enhanced iron core acoustic pattern is input, the operating feature recognizer can automatically identify whether there are characteristics such as abnormal low-frequency vibration frequency and enhanced energy in a specific frequency band that are unique to iron core loosening in the pattern, and then extract the corresponding iron core operating features. If the pattern is found to match the feature pattern of slight loosening, the operating feature of "slight iron core loosening" is output, providing a basis for subsequent anomaly judgment.

[0112] Through the above steps, the original core acoustic signature signal is transformed into a distinctive and interference-resistant operating feature. This not only solves the problem of indistinct features in traditional acoustic signature analysis, but also provides reliable input data for subsequent anomaly identification, effectively improving the accuracy and robustness of core loosening fault diagnosis.

[0113] S130: Identify core loosening anomalies based on the core operating characteristics to obtain core loosening anomaly characteristics;

[0114] In this embodiment of the application, in order to accurately distinguish between the normal operation and loose state of the iron core, it is necessary to compare and analyze the preset normal operation feature template and operation features, and combine the abnormal threshold to determine the state of the iron core and extract abnormal features, so as to clarify whether the iron core has a loose problem.

[0115] Specifically, it is first necessary to collect historical data on the transformer core under normal operating conditions in advance, and then construct a normal operating characteristic template for the core based on this historical data. During the construction process, it is necessary to ensure that the historical data covers all types of operating conditions under normal core operation, so that the normal characteristic template can fully reflect the typical operating state of the core when it is not loose.

[0116] Furthermore, the core operating characteristics output by the operating feature recognizer will be comprehensively compared and analyzed with the constructed core normal operation characteristic template. The comparison will focus on key feature dimensions that reflect core loosening, thereby obtaining the operating characteristic deviation value.

[0117] Furthermore, the obtained operational characteristic deviation is compared with a preset abnormal threshold. When the operational characteristic deviation exceeds the preset abnormal threshold, it can be determined that the core is loose. The parts that are significantly different from the normal characteristics are extracted from the current core operational characteristics. These differences are the core loosening abnormal characteristics, which contain feature information that reflects the degree and location of loosening.

[0118] This step, by comparing real-time operating characteristics with normal templates and combining threshold judgments, clearly distinguishes between normal and loose core states and extracts abnormal features. This provides an analytical basis for subsequent verification of the authenticity of the anomaly through winding electrical parameters, avoids diagnostic bias caused by relying solely on a single feature, and further improves the accuracy of core loosening fault diagnosis.

[0119] Step S130 in the method provided in this application embodiment includes:

[0120] Collect historical normal operation feature sets of transformer cores under normal conditions, and establish a normal operation feature template of the core based on the historical normal operation feature sets of the cores;

[0121] The operating characteristics of the iron core are compared and analyzed with the normal operating characteristics template of the iron core to calculate the operating characteristic deviation;

[0122] When the deviation of the operating characteristics exceeds the preset abnormal threshold, the operating characteristics of the iron core are regarded as abnormal characteristics of iron core loosening.

[0123] In this embodiment of the application, in order to avoid misjudgment or omission of abnormalities due to the lack of clear normal benchmarks, it is necessary to realize the identification of core loosening abnormalities by constructing a normal operation feature template, comparing real-time operation features with the template, and combining threshold judgment, so as to extract abnormal features that can clearly reflect the core loosening problem and ensure the accuracy of the diagnostic results.

[0124] Specifically, the first step is to collect a historical set of normal operating characteristics of the transformer core under normal conditions. During the data collection process, it is necessary to cover various typical operating conditions of the core, such as normal operating states under different load intensities, different ambient temperatures, and different operating durations, to ensure that the historical set of normal operating characteristics comprehensively reflects the operating characteristics of the core when it is not loose.

[0125] For example, the acoustic signature features of the transformer core under different load conditions such as rated load, 50% load, and light load are collected when the core does not show signs of loosening, as well as the stable operating parameter correlation features under the corresponding operating conditions. This ensures that the historical core normal operation feature set contains a sufficiently rich set of normal state samples, avoiding the inapplicability of subsequent templates to actual operating scenarios due to the limited number of samples.

[0126] Furthermore, a core normal operation characteristic template is established based on the obtained historical core normal operation characteristic set. During the establishment process, statistical analysis and integration of various characteristic data from the historical characteristic set are required. For example, mean calculation and variance analysis are performed on data such as the normal acoustic frequency distribution and characteristic amplitude range under different operating conditions to determine the reasonable fluctuation range of each characteristic under normal conditions. The reasonable fluctuation range and typical characteristic values ​​are then integrated into a structured template to ensure that the template clearly defines the characteristic boundaries of core normal operation.

[0127] For example, if the historical core normal operation feature set includes normal acoustic signature data of the target transformer under three operating conditions: rated load, 50% load, and light load, the frequency distribution analysis of these data calculates that the main frequency range of normal acoustic signature under rated load is 50-150Hz, with a mean of 100Hz and a variance of 15Hz. 2 At 50% load, the main frequency range is 45-140Hz, with a mean of 95Hz and a variance of 12Hz. 2 Under light load, the main frequency range is 40-130Hz, with a mean of 90Hz and a variance of 10Hz. 2 .

[0128] During the characteristic amplitude analysis, the maximum fluctuation range of the normal voiceprint amplitude under the above three operating conditions was found to be 0.1-0.5V. The obtained main frequency ranges, mean, variance, and amplitude ranges were then integrated to form a structured template containing characteristic boundaries under different operating conditions. In subsequent comparisons, if the real-time voiceprint frequency exceeds the 40-150Hz range and the amplitude exceeds 0.1-0.5V, a preliminary judgment can be made that an anomaly may exist.

[0129] Furthermore, the core operating characteristics output by the operating feature recognizer are compared and analyzed with the constructed core normal operation feature template to calculate the operating feature deviation.

[0130] Specifically, the comparison should focus on key characteristic dimensions that directly reflect core loosening, such as the amplitude of low-frequency vibration characteristics in the core acoustic signal and the energy proportion of specific frequency bands. The operational characteristic deviation is obtained by calculating the numerical differences between the real-time operational characteristics and the corresponding characteristics in the core's normal operation characteristic template, and the degree of deviation from the reasonable range.

[0131] For example, the deviation result can reflect the degree of difference between the real-time low-frequency vibration characteristic amplitude exceeding the upper limit of normal fluctuation in the core normal operation characteristic template, or the percentage difference between the energy ratio of a specific frequency band and the template mean.

[0132] Finally, the calculated operating characteristic deviation is compared with the preset abnormal threshold. The preset abnormal threshold needs to be determined based on historical data of a large number of core loosening fault samples. For example, by analyzing the changing pattern of operating characteristic deviation of multiple transformers of the same model from slight loosening to severe loosening, a threshold is set that can effectively identify slight loosening while avoiding false triggering of normal operating condition fluctuations.

[0133] Specifically, when the deviation of the operating characteristics exceeds the preset abnormal threshold, it indicates that the current operating state of the iron core has deviated from the normal range and there is a loosening problem. At this time, the current iron core operating characteristics are taken as the iron core loosening abnormal characteristics, which include information such as frequency and amplitude that exceed the normal range.

[0134] For example, if, for a certain type of transformer, by analyzing fault samples of 10 devices of the same type ranging from slight loosening to severe loosening, it is found that when the loosening is slight, the deviation of the operating characteristics exceeds the upper limit of normal fluctuation for the first time by 1.5 times, and when the loosening is severe, the deviation reaches 3 times the upper limit of normal, and the maximum fluctuation deviation under normal operating conditions is only 1.2 times the upper limit of normal, then the preset abnormal threshold is set to 1.5 times the upper limit of normal fluctuation.

[0135] Furthermore, when the calculated deviation of the real-time operating characteristics of a target transformer is 1.8 times the normal upper limit, exceeding the preset threshold of 1.5 times, it can be determined that the core is loose. At this time, the real-time operating characteristics are regarded as abnormal characteristics of core looseness, which include information such as 200-250Hz frequency components (exceeding the normal 50-150Hz range) and 0.7V amplitude (exceeding the normal 0.1-0.5V range).

[0136] Through the above steps, a complete process from establishing normal benchmarks to identifying anomalies and extracting abnormal features is realized, providing a clear basis for identifying core loosening anomalies and avoiding the limitations of relying solely on experience or single feature judgments.

[0137] S140: Verify the abnormal characteristics of the loose core based on the set of winding electrical parameters, and issue a loose core fault signal according to the verification results.

[0138] In this embodiment of the application, in order to avoid false alarms caused by relying solely on the acoustic signature monitoring of a single iron core and to improve the reliability of the iron core loosening diagnosis results, it is necessary to combine the synchronously collected winding electrical parameters to conduct collaborative verification of abnormal features, so as to determine whether the iron core does indeed have a loosening fault through cross-confirmation of multi-source data.

[0139] Specifically, the abnormal timestamp intervals corresponding to the previously identified core loosening abnormality characteristics are first obtained. Based on the abnormal timestamp intervals, electrical parameters that completely correspond to the time period are extracted from the synchronously collected winding electrical parameter set to obtain the set of electrical parameters to be verified, so as to ensure that the extracted winding electrical parameters completely match the abnormal core time period.

[0140] Furthermore, feature extraction is performed on the set of electrical parameters to be verified to obtain the electrical features of the winding to be verified. During the extraction process, it is necessary to focus on key parameter features that can reflect the impact of core anomalies on the winding. By processing the key parameter features, the original electrical parameters are transformed into feature data that can be directly used for comparative analysis, making the correlation between the winding electrical parameters and core anomalies easier to demonstrate.

[0141] Furthermore, a winding response analyzer is retrieved. This analyzer is pre-trained based on historical core loosening records of transformers of the same model and has the ability to predict the winding response based on the abnormal characteristics of core loosening. The abnormal core loosening characteristics are input into the winding response analyzer, and through the analyzer's calculations, the theoretically expected electrical characteristics of the winding response when the core exhibits this type of loosening abnormality are obtained, serving as a benchmark reference characteristic for verification.

[0142] Furthermore, the similarity between the electrical characteristics of the winding to be verified and the electrical characteristics of the winding response is calculated to obtain the electrical characteristic similarity, which is then used as the verification result.

[0143] Finally, the obtained electrical feature similarity is compared with a preset similarity threshold. When the electrical feature similarity is greater than or equal to the preset similarity threshold, it indicates that the actual electrical characteristics of the winding highly match the expected response characteristics when the iron core is loose, and the iron core loosening anomaly truly exists. At this time, an iron core loosening fault signal is issued, providing a clear basis for subsequent operation and maintenance decisions.

[0144] This step verifies the abnormal characteristics of core loosening by using winding electrical parameters, forming a dual-layer diagnostic method of core acoustic monitoring and winding verification. This effectively compensates for the limitations of a single monitoring method and further improves the accuracy and reliability of core loosening fault diagnosis.

[0145] Step S140 in the method provided in this application embodiment includes:

[0146] Obtain the abnormal timestamp interval of the core loosening abnormality feature, and extract the corresponding electrical parameters from the winding electrical parameter set based on the abnormal timestamp interval to obtain the electrical parameter set to be verified;

[0147] Feature extraction is performed on the set of electrical parameters to be verified to obtain the electrical characteristics of the winding to be verified;

[0148] The winding response analyzer is invoked, and the abnormal characteristics of the loose core are processed based on the winding response analyzer to obtain the winding response electrical characteristics.

[0149] Calculate the similarity between the electrical characteristics of the winding to be verified and the electrical characteristics of the winding response, and obtain the electrical characteristic similarity as the verification result;

[0150] When the electrical feature similarity is greater than or equal to a preset similarity threshold, a core loosening fault signal is issued.

[0151] In this embodiment of the application, in order to improve the reliability of the core loosening diagnosis results through multi-source data cross-verification, it is necessary to combine the synchronously collected winding electrical parameters to jointly verify the identified core loosening abnormal features, so as to clarify the authenticity of the abnormality and ensure that the final fault signal can provide an accurate basis for power system operation and maintenance.

[0152] Specifically, firstly, the abnormal timestamp interval of the core loosening anomaly is obtained. This abnormal timestamp interval records the specific start and end times when the core is determined to be loose. Simultaneously, based on this abnormal timestamp interval, the winding electrical parameters for the corresponding time period are extracted from the synchronously collected winding electrical parameter set to obtain the set of electrical parameters to be verified.

[0153] For example, if the abnormal timestamp interval corresponding to the abnormal feature of loose iron core is 14:20:00-14:20:30, the winding electrical parameters such as excitation current harmonics and three-phase current imbalance within this 30-second period are extracted from the winding electrical parameter set to form a set of electrical parameters to be verified.

[0154] Furthermore, feature extraction is performed on the obtained set of electrical parameters to be verified to obtain the electrical features of the winding to be verified, so as to transform the original electrical parameters into feature data that can be directly used for comparative analysis.

[0155] Specifically, the extraction process needs to focus on key parameters that can reflect the impact of core anomalies on winding conduction. For the excitation current harmonic data in the electrical parameter set to be verified, the amplitude ratio of each harmonic relative to the fundamental wave is calculated to clarify the strength distribution of the harmonic components.

[0156] In addition, for the three-phase current imbalance data, the peak value, average value and fluctuation amplitude of the imbalance within a specific time period are calculated to capture the imbalance characteristics of the current distribution; if the parameter set includes winding vibration signals, it is also necessary to extract the main frequency amplitude, frequency bandwidth and other characteristics of the vibration signals.

[0157] Furthermore, after obtaining the electrical characteristics of the winding to be verified, the winding response analyzer is invoked, and the core loosening anomaly characteristics are correlated and mapped to obtain the winding response electrical characteristics, so as to obtain the electrical characteristic benchmark that the winding should theoretically present when the core exhibits this type of loosening anomaly.

[0158] In the method provided in this application embodiment, the construction steps of the winding response analyzer include:

[0159] Obtain the transformer model of the target transformer, and retrieve multiple historical core loosening anomaly records of transformers of the same model based on the transformer model. Each core loosening anomaly record includes historical core loosening anomaly characteristics and historical winding electrical characteristics.

[0160] A sample core loosening anomaly feature set is constructed based on the historical core loosening anomaly features in multiple historical core loosening anomaly records, and a sample winding response electrical feature set is constructed based on the historical winding electrical features in multiple historical core loosening anomaly records;

[0161] Using the sample core loosening anomaly feature set as input features and the sample winding response electrical feature set as output target, the winding response analyzer is trained and generated.

[0162] Specifically, the first step is to obtain the transformer model of the target transformer. This model information is directly related to key characteristics such as the transformer's core structure, number of winding turns, and rated parameters. Based on the obtained transformer model, historical records of core loosening anomalies from multiple transformers that perfectly match the target model are retrieved from the transformer operation and maintenance database.

[0163] Each historical record of core loosening anomaly must include two core pieces of information: First, the characteristics of the historical core loosening anomaly, such as the frequency range and amplitude fluctuation data of the abnormal frequency band in the core acoustic signature signal, or the processed abnormal acoustic signature spectrum characteristics. Second, the corresponding historical winding electrical characteristics, including parameters such as the proportion of harmonic amplitude of the winding excitation current, the value of the three-phase current imbalance, and the main frequency of the winding vibration signal at the time the core loosening occurred.

[0164] At the same time, during the retrieval process, it is necessary to ensure that there is a sufficient number of historical records. For example, at least 30 or more records of core loosening anomalies of the same type of transformer should be collected, while covering cases with different degrees of loosening, so as to ensure that the sample set constructed later can fully reflect the correlation between core loosening and winding electrical response.

[0165] After obtaining a sufficient number of historical records of core loosening anomalies, a sample core loosening anomaly feature set is constructed based on the historical core loosening anomaly features in the records. During the construction process, each historical anomaly feature needs to be standardized. For example, the frequency data of the acoustic anomaly frequency band in different records is uniformly converted into a ratio relative to the fundamental frequency, and the amplitude fluctuation data is normalized to the same order of magnitude to ensure that the format of the feature data in the sample set is consistent.

[0166] Simultaneously, a sample winding response electrical feature set is constructed based on the historical winding electrical characteristics in historical records of core loosening anomalies. Similarly, the historical winding electrical characteristics need to be standardized. For the excitation current harmonic characteristics, the amplitude ratio of each harmonic relative to the fundamental wave is calculated as a unified feature index; for the three-phase current imbalance, the calculation method specified by national standards is used for unified quantification; for the winding vibration signal, the dominant vibration frequency, peak amplitude, and duration are extracted as core features.

[0167] Through the standardization process described above, each data point in the electrical feature set of the sample winding response can be accurately associated with the corresponding record in the feature set of sample core loosening anomalies.

[0168] Furthermore, after completing the construction of the sample core loosening anomaly feature set and the sample winding response electrical feature set, a deep learning network architecture was selected to build the basic model of the winding response analyzer.

[0169] Specifically, a multilayer perceptron (MLP) is chosen as the basic network architecture. This architecture can effectively handle the mapping relationship of multi-dimensional features and adapt to the nonlinear correlation learning from iron core anomaly features to winding electrical features.

[0170] During the architecture construction process, network parameters need to be reasonably set according to the size and feature dimensions of the sample set. For example, if the sample set contains 50 records, the feature dimension of the sample core loosening anomaly feature set is 8-dimensional, and the output dimension of the sample winding response electrical feature set is 4-dimensional, then an MLP architecture with 3 hidden layers can be constructed. The input layer has 8 neurons to match the input feature dimension, the first hidden layer has 64 neurons, the second hidden layer has 32 neurons, the third hidden layer has 16 neurons, and the output layer has 4 neurons corresponding to the output target dimension. Each hidden layer uses the ReLU activation function to enhance the nonlinear fitting ability, and the output layer uses a linear activation function to adapt to the continuous numerical output requirements of electrical parameters.

[0171] Furthermore, the MLP model is trained under supervision by using the sample core loosening anomaly feature set as input features and the sample winding response electrical feature set as output target.

[0172] Before actual training, the sample set is first divided into a training set, a validation set, and a test set in a ratio of 7:2:1. The training set is used for iterative updates of model parameters, the validation set is used to monitor model performance during training to avoid overfitting, and the test set is used to finally evaluate the model's generalization ability.

[0173] During training, mean squared error (MSE) was used as the loss function to quantify the deviation between the model's predicted winding electrical characteristics and the actual winding electrical characteristics of the samples. The model parameters were adjusted using the Adam optimizer, with an initial learning rate of 0.001 and a batch size of 8. After every 10 training rounds, the model's prediction accuracy was tested using validation set data, and the mean absolute error (MAE) between the predicted and actual values ​​was calculated. If the MAE no longer decreased or showed an upward trend for 5 consecutive rounds, it indicated that the model had converged, and training was stopped.

[0174] Simultaneously, periodic validation and parameter adjustments are necessary during training. For example, if the MAE on the validation set rises from 0.02 to 0.035 after a certain training round, indicating an overfitting trend, the learning rate can be reduced to 0.0005 to suppress overfitting. If the model's prediction error for a certain type of sample is found to be significantly higher than that for other types of samples, multiple historical records of that type are added to the sample set, the training and validation sets are re-split, and training continues until the MAE on the validation set stabilizes below 0.02, and the difference between the MAE on the test set and the MAE on the validation set is less than 0.005, ensuring that the model has stable predictive ability and good generalization.

[0175] Finally, the winding response analyzer trained through the above steps can accurately output the winding response electrical characteristics of the same type of transformer under this type of anomaly based on the input core loosening anomaly characteristics.

[0176] Furthermore, after the winding response analyzer is constructed, when an abnormal feature of core loosening is identified through the core acoustic signature signal, the winding response analyzer can be invoked, and the currently acquired abnormal feature of core loosening can be input into the analyzer. At this time, the winding response analyzer will perform feature mapping and parameter calculation processing on the input abnormal feature based on the correlation law between core loosening and winding electrical response of the same type of transformer learned during the training process.

[0177] For example, the abnormal features in the core acoustic waveform, such as an abnormal increase of 15% in amplitude in the 200-250Hz frequency band and an increase in spectral entropy to 0.8, are mapped to the corresponding predicted values ​​of winding electrical parameters. Finally, the winding response electrical features are output. These features clearly define the theoretical electrical state benchmark that the winding should present when the core exhibits the current type of loosening abnormality, such as 8% of the second harmonic of the excitation current, 5% of the third harmonic, 0.06 of the three-phase current imbalance, and 180Hz of the winding vibration main frequency.

[0178] Furthermore, after obtaining the electrical characteristics of the winding response, it is necessary to calculate the similarity between the electrical characteristics of the winding to be verified and the electrical characteristics of the response, so as to obtain the electrical characteristic similarity as the verification result.

[0179] Specifically, the cosine similarity calculation method is used to quantify the similarity of electrical features. First, the key dimension parameters of the electrical features of the winding to be verified and the electrical features of the winding response are organized into two feature vectors of the same dimension. Then, the cosine value of the angle between the two vectors is calculated by the cosine similarity formula.

[0180] The cosine value ranges from [-1, 1]. The closer the value is to 1, the higher the matching degree between the two types of features. The closer it is to 0, the lower the matching degree. The final cosine value is the electrical feature similarity, which is used to quantitatively judge the consistency between the electrical features of the winding to be verified and the expected response features.

[0181] Furthermore, the obtained electrical feature similarity is compared with a preset similarity threshold to determine the authenticity of the core loosening abnormality and output the corresponding diagnostic signal.

[0182] The preset similarity threshold needs to be determined by combining a large amount of historical verification data of transformers of the same model. For example, by analyzing the similarity distribution of multiple sets of real core loosening cases and false anomaly cases caused by interference, 90% is set as the similarity threshold. This similarity threshold can ensure that when a real loosening fault occurs, the similarity between the actual winding electrical characteristics and the expected response characteristics meets the standard, and can also eliminate false anomalies with low similarity caused by factors such as power grid fluctuations and temporary sensor interference.

[0183] Specifically, when the electrical feature similarity is greater than or equal to the preset similarity threshold, it indicates that the actual electrical state of the current winding is highly matched with the response characteristics that the core loosening abnormality should have, and the core loosening abnormality is real. At this time, a core loosening fault signal is issued to prompt the operation and maintenance personnel to repair the transformer core in a timely manner.

[0184] In the method provided in this application embodiment, when the electrical feature similarity is less than a preset similarity threshold, an abnormal core state signal is issued.

[0185] Specifically, when the electrical feature similarity is less than the preset similarity threshold, it indicates that the actual electrical state of the current winding does not exhibit the response characteristics that should be present in an abnormal core loosening. The abnormality identified by the core acoustic signal may be a false abnormality caused by external interference or non-loosening factors, rather than a true core loosening. In this case, an abnormal core status signal is issued. This signal does not directly determine a core loosening fault, but rather prompts maintenance personnel to further investigate the cause of the abnormality.

[0186] For example, check whether there are interference sources in the soundprint sensor's acquisition environment, check whether the winding electrical parameter acquisition equipment is operating normally, and compare the transformer's other operating data during the same period to see if it is stable. By checking from multiple dimensions, the cause of the abnormality can be confirmed, so as to avoid starting unnecessary maintenance procedures due to misjudgment, and also to prevent the omission of potential non-loose equipment problems.

[0187] Ultimately, by extracting electrical parameters to be verified based on abnormal timestamps, transforming feature extraction into comparable winding electrical features, and then using a winding response analyzer trained on the same model to obtain expected benchmarks, calculate cosine similarity, and combine it with threshold judgment, a two-layer diagnostic approach of identifying core acoustic fingerprint anomalies and verifying winding electrical parameters was achieved. This effectively improves the accuracy and reliability of identifying transformer core loosening faults, avoids the waste of maintenance resources and equipment safety risks caused by misjudgment or omission, and ensures the long-term stable operation of transformers.

[0188] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects:

[0189] This application proposes a method for diagnosing transformer core loosening based on acoustic signature mapping and data enhancement. First, acoustic signature sensors are deployed at multiple key locations on the transformer core to form an array, and acoustic signature signals are collected at a preset sampling frequency. Simultaneously, preset winding electrical parameters are acquired, and a monitoring unit is configured for the windings to synchronously collect these parameters at the same preset sampling frequency, resulting in a set of core acoustic signature signals and winding electrical parameters. Next, a preset judgment window is set to extract segments of the acoustic signature signal. The short-time Fourier transform window length is dynamically adjusted based on the instantaneous signal-to-noise ratio and frequency change rate of the acoustic signature signal segment. The signal is then converted into a two-dimensional acoustic signature map using a trainable Mel filter bank, and finally weighted by spectral entropy. Random time-frequency occlusion enhancement processing is used to obtain an enhanced core acoustic signature, which is then input into an operating feature recognizer to extract core operating features. Subsequently, historical data of normal core operation is collected to construct a core normal operation feature template. The operating feature deviation is calculated by comparing it with real-time operating features. When the operating feature deviation exceeds a preset threshold, core loosening abnormal features are extracted. Finally, the corresponding winding electrical parameters are extracted based on the abnormal timestamp, and features to be verified are extracted. A winding response analyzer trained based on historical data of the same type of transformer is called to obtain the expected response features. The cosine similarity between the two types of features is calculated, and a core loosening fault signal or core state abnormality signal is issued according to a preset similarity threshold.

[0190] The method provided in this application, through a technical solution of "synchronous acquisition of multi-source data - acoustic signature enhancement and feature recognition - normal template comparison for anomaly detection - winding parameter collaborative verification," solves the problems of false alarms in traditional single monitoring methods, difficulty in extracting key features from complex core acoustic signature signals, and lack of multi-source verification support for diagnostic results. This effectively improves the accuracy and reliability of transformer core loosening fault diagnosis, providing scientific technical support for transformer operation and maintenance in power systems.

[0191] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0192] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0193] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A transformer core looseness diagnosis method based on voiceprint map and data enhancement, characterized in that, The method comprises: Collecting a voiceprint signal of a transformer core of a target transformer to obtain a core voiceprint signal, and synchronously collecting electrical parameters of a transformer winding of the target transformer to obtain a winding electrical parameter set; Cutting a voiceprint signal segment from the core voiceprint signal, performing data enhancement processing on the voiceprint signal segment to obtain an enhanced core voiceprint atlas, and performing core operation feature recognition according to the enhanced core voiceprint atlas to obtain a core operation feature; Performing core loosening abnormality recognition according to the core operation feature to obtain a core loosening abnormality feature; Verifying the core loosening abnormality feature based on the winding electrical parameter set, and issuing a core loosening fault signal according to a verification result; Verifying the core loosening abnormality feature based on the winding electrical parameter set, and issuing a core loosening fault signal according to a verification result, comprising: Obtaining an abnormal timestamp interval of the core loosening abnormality feature, extracting corresponding electrical parameters in the winding electrical parameter set based on the abnormal timestamp interval to obtain a to-be-verified electrical parameter set; Performing feature extraction on the to-be-verified electrical parameter set to obtain a to-be-verified winding electrical feature; Calling a winding response analyzer, processing the core loosening abnormality feature based on the winding response analyzer to obtain a winding response electrical feature; Calculating a similarity between the to-be-verified winding electrical feature and the winding response electrical feature to obtain an electrical feature similarity as the verification result; When the electrical feature similarity is greater than or equal to a preset similarity threshold, issuing a core loosening fault signal.

2. The method of claim 1, wherein, Collecting a voiceprint signal of a transformer core of a target transformer to obtain a core voiceprint signal, comprising: Arranging voiceprint sensors at multiple positions of the transformer core to form a voiceprint sensor array; Collecting a voiceprint signal of the transformer core through the voiceprint sensor array at a preset sampling frequency to obtain the core voiceprint signal.

3. The method of claim 2, wherein, Synchronously collecting electrical parameters of a transformer winding of the target transformer to obtain a winding electrical parameter set, comprising: Obtaining a preset winding electrical parameter, and configuring a winding electrical parameter monitoring unit for the transformer winding based on the preset winding electrical parameter; Collecting electrical parameters of the transformer winding through the winding electrical parameter monitoring unit at the preset sampling frequency to obtain the winding electrical parameter set.

4. The method of claim 3, wherein, The preset winding electrical parameter comprises at least one of an excitation current harmonic, a three-phase current unbalance degree, and a winding vibration signal.

5. The method of claim 1, wherein, Cutting a voiceprint signal segment from the core voiceprint signal, performing data enhancement processing on the voiceprint signal segment to obtain an enhanced core voiceprint atlas, comprising: Setting a preset judgment window, and cutting a voiceprint signal segment from the core voiceprint signal based on the preset judgment window; Dynamically adjusting a window length parameter of short-time Fourier transform according to an instantaneous signal-to-noise ratio and a frequency change rate of the voiceprint signal segment, performing time-frequency analysis on the voiceprint signal segment to obtain time-frequency domain data; Performing filter processing on the time-frequency domain data through a trainable Mel filter bank to convert into a two-dimensional voiceprint atlas; The two-dimensional voiceprint spectrum is subjected to enhancement processing to obtain an enhanced core voiceprint spectrum, the enhancement processing including weighted processing based on spectral entropy and random time-frequency occlusion processing.

6. The method of claim 5, wherein, Core operation feature recognition is performed according to the enhanced core voiceprint spectrum to obtain a core operation feature, including: An operation feature recognizer is called, the operation feature recognizer being constructed based on a sample enhanced core voiceprint spectrum set and a sample core operation feature set; The enhanced core voiceprint spectrum is input into the operation feature recognizer for core operation feature recognition, and a core operation feature is output.

7. The method of claim 1, wherein, Core looseness abnormality recognition is performed according to the core operation feature to obtain a core looseness abnormality feature, including: A historical core normal operation feature set of the transformer core in a normal state is collected, and a core normal operation feature template is established according to the historical core normal operation feature set; The core operation feature is compared and analyzed with the core normal operation feature template, and an operation feature deviation is calculated; When the operation feature deviation exceeds a preset abnormality threshold, the core operation feature is taken as a core looseness abnormality feature.

8. The method of claim 1, wherein, The construction steps of the winding response analyzer include: A transformer model of the target transformer is obtained, and a plurality of historical core looseness abnormality records of the same model transformer are retrieved according to the transformer model, each core looseness abnormality record including a historical core looseness abnormality feature and a historical winding electrical feature; A sample core looseness abnormality feature set is constructed based on the historical core looseness abnormality features in the plurality of historical core looseness abnormality records, and a sample winding response electrical feature set is constructed based on the historical winding electrical features of the plurality of historical core looseness abnormality records; The sample core looseness abnormality feature set is taken as an input feature, and the sample winding response electrical feature set is taken as an output target to train and generate the winding response analyzer.

9. The method of claim 1, wherein, When the electrical feature similarity is less than a preset similarity threshold, a core state abnormality signal is issued.

Images