Method for detecting acoustic anomalies of transformer
The transformer acoustic signature anomaly detection method, which utilizes a multi-sensor layout and deep learning feature fusion, solves the problem of low efficiency in transformer fault detection in existing technologies, achieves efficient and accurate anomaly detection, and improves the monitoring capability of transformer operating status.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- SHANGHAI SHANGPENG ELECTRIC CO LTD
- Filing Date
- 2025-07-22
- Publication Date
- 2026-06-25
AI Technical Summary
Existing transformer fault detection methods rely on regular maintenance and manual inspection, which are inefficient and make it difficult to detect potential faults in a timely manner. Existing voiceprint-based anomaly detection methods are insufficient in feature extraction and anomaly detection, making it difficult to meet the needs of practical applications.
Multiple high-sensitivity acoustic signature sensors are used to collect the acoustic signature signal of the transformer. The signal is then denoised and segmented using an adaptive filter. By combining time-domain, spectral, and deep learning feature extraction, a support vector machine model is used for anomaly detection. An attention mechanism is also introduced to form a comprehensive feature vector for judgment.
This improves the efficiency and accuracy of transformer operation status monitoring, enhances the accuracy and robustness of detection, reduces the possibility of missed detections and false detections, and ensures the comprehensiveness and reliability of detection results.
Smart Images
Figure CN2025109741_25062026_PF_FP_ABST
Abstract
Description
A method for detecting abnormal sound signatures in transformers Technical Field
[0001] This invention relates to the field of transformer technology, specifically a method for detecting abnormal sound signatures in transformers. Background Technology
[0002] As a critical piece of equipment in the power system, the operating status of transformers directly affects the stability and security of the entire power system. However, traditional transformer fault detection methods mainly rely on periodic maintenance and manual inspection, which are inefficient and make it difficult to detect potential faults in a timely manner. In recent years, with the development of acoustic signature analysis technology, acoustic signature-based transformer anomaly detection methods have gradually attracted attention. However, existing methods still have shortcomings in feature extraction and anomaly detection, making it difficult to meet the needs of practical applications. Therefore, a transformer acoustic signature anomaly detection method based on deep learning and time-frequency domain joint analysis is needed. Through advanced feature extraction, time-frequency domain joint analysis, and anomaly detection model optimization, this method can achieve efficient and accurate monitoring of the transformer's operating status. Summary of the Invention
[0003] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.
[0004] To achieve the above objectives, the present invention provides the following technical solution:
[0005] A method for detecting transformer acoustic anomalies includes: using multiple high-sensitivity acoustic sensors to collect acoustic signals during transformer operation; preprocessing the collected acoustic signals; extracting time-domain features, frequency-domain features, and deep-level features of the acoustic signals; fusing the time-domain features, frequency-domain features, and deep learning features to form a comprehensive feature vector; constructing an anomaly detection model using a support vector machine and introducing an attention mechanism into the model; inputting the comprehensive feature vector into the anomaly detection model for detection to obtain the detection result; and recording each detection result.
[0006] As a further aspect of the present invention: the method of using multiple high-sensitivity acoustic signature sensors to collect acoustic signature signals during transformer operation involves arranging multiple high-sensitivity acoustic signature sensors around the transformer, with at least four sensors distributed at the four corners of the transformer, as expressed in the following formula:
[0007] ;
[0008] in, To optimize the sensor layout, For the number of sensors, For the first The weight of each sensor, For the first Sensor sensitivity;
[0009] A high-sensitivity MEMS microphone was selected as the voiceprint sensor, with a frequency response range between 20Hz and 20kHz; a high sampling rate data acquisition device was used, with a sampling rate of no less than 20kHz.
[0010] As a further aspect of the present invention: the preprocessing of the acquired voiceprint signal specifically involves using an adaptive filter to remove background noise from the voiceprint signal, expressed as:
[0011] ;
[0012] in, This is the denoised voiceprint signal. This is the original voiceprint signal. As a smoothing factor, This is the estimated value of the denoised signal from the previous time step;
[0013] The collected voiceprint signals are divided into segments of fixed length, with each segment ranging from 1 to 5 seconds in length.
[0014] As a further aspect of the present invention: the extraction of time-domain features, spectral features, and deep-level features of the voiceprint signal specifically involves the following steps: calculating the time-domain features of each voiceprint segment, including mean, variance, zero-crossing rate, and energy; converting the voiceprint signal into a frequency-domain representation using a short-time Fourier transform and extracting spectral features; and inputting the time-domain features and spectral features into a convolutional neural network and a recurrent neural network to automatically extract deep-level features, expressed as:
[0015] ;
[0016] ;
[0017] in, For time-domain feature vectors, For spectral feature vectors, It is a convolutional neural network. It is a recurrent neural network. These are deep-seated features.
[0018] As a further aspect of the present invention: the fusion of time-domain features, frequency-domain features, and deep learning features to form a comprehensive feature vector specifically involves fusing the extracted time-domain features, frequency-domain features, and deep learning features to form a comprehensive feature vector, expressed as:
[0019] ;
[0020] in, For the comprehensive feature vector, and This is the weighting factor.
[0021] As a further aspect of the present invention: the anomaly detection model is constructed using a support vector machine (SVM), and an attention mechanism is introduced into the model. Specifically, the steps are as follows: A support vector machine is selected as the anomaly detection model, and an attention mechanism is introduced into the SVM model to enable the model to focus on the features most helpful for anomaly detection. The expression is:
[0022] ;
[0023] ;
[0024] in, Let be the decision function. The number of support vectors, For Lagrange multipliers, For the first The label of each sample For kernel function, Bias term, For attention weights, For feature dimension, For the first The weights of each feature, For the first The values of each feature.
[0025] As a further aspect of the present invention: the step of inputting the comprehensive feature vector into the anomaly detection model for detection to obtain the detection result is as follows: the comprehensive feature vector is input into the anomaly detection model for detection. The input is fed into the pre-trained support vector machine model; the result is determined as an anomaly based on the anomaly score, using the following expression:
[0026] ;
[0027] in, For the test results, This is the anomaly threshold; if the anomaly score is greater than the threshold... If it is, it is judged as abnormal; otherwise, it is judged as normal.
[0028] As a further aspect of the present invention: the specific steps for recording each detection result are as follows: creating a structured data table to record the results of each detection; using a CSV file or database table to store the detection results; and recording the detection results into the data table after each detection is completed.
[0029] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the transformer acoustic anomaly detection method as described in the first aspect of the present invention.
[0030] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is processed.
[0031] Compared with the prior art, the beneficial effects of the present invention are:
[0032] By employing a multi-sensor layout and a high-sensitivity MEMS microphone, high resolution and high signal-to-noise ratio of the acoustic signature signal are ensured. The multi-sensor arrangement can cover different locations of the transformer, thereby capturing the transformer's operating status more comprehensively. The extraction of time-domain and spectral features describes the characteristics of the acoustic signature signal from different perspectives, while convolutional neural networks and recurrent neural networks can automatically extract deep-level features and capture complex signal patterns. Multi-dimensional feature extraction methods can comprehensively describe the characteristics of the acoustic signature signal, improving the richness and diversity of features. The extraction of deep-level features can capture the implicit patterns of the signal, providing rich feature support for subsequent anomaly detection. The feature fusion step ensures the comprehensiveness and representativeness of the comprehensive feature vector, improving the quality of the model's input data. The comprehensive feature vector contains information from the time domain, frequency domain, and deep-level features, which can better reflect the transformer's operating status. Attached Figure Description
[0033] Figure 1 is a flowchart illustrating a method for detecting abnormal acoustic signatures in transformers. Detailed Implementation
[0034] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0035] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.
[0036] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.
[0037] Example
[0038] Please refer to Figure 1, which illustrates the first embodiment of the present invention. This embodiment provides a method for detecting abnormal acoustic signatures in transformers, comprising:
[0039] S1. Multiple high-sensitivity acoustic sensors are used to collect acoustic signals during transformer operation;
[0040] Specifically, four high-sensitivity MEMS microphones are arranged around the transformer, located at the four corners of the transformer. The frequency response range of the sensors is 20Hz to 20kHz, and the sampling rate is no less than 20kHz.
[0041] It should be noted that the sensor arrangement can cover different locations of the transformer, thereby capturing the transformer's operating status more comprehensively. The high-sensitivity MEMS microphone and high-sampling-rate data acquisition equipment ensure high resolution and high signal-to-noise ratio of the voiceprint signal, providing a high-quality data foundation for subsequent feature extraction and anomaly detection.
[0042] S2. Preprocess the acquired voiceprint signals;
[0043] Specifically, an adaptive filter is used to remove background noise from the voiceprint signal, and the acquired voiceprint signal is divided into segments of fixed length, each segment being 2 seconds long.
[0044] It should be noted that the adaptive filter can effectively remove background noise and improve the signal-to-noise ratio of the voiceprint signal, while the signal segmentation divides the continuous voiceprint signal into segments of fixed length, which facilitates subsequent feature extraction and processing. At the same time, it balances computational complexity and feature integrity, ensuring that each segment contains sufficient information.
[0045] S3. Extract the time-domain features, spectral features, and deep-level features of the voiceprint signal;
[0046] Specifically, the temporal features of each voiceprint segment are calculated, including mean, variance, zero-crossing rate, and energy. The voiceprint signal is converted into a frequency domain representation using short-time Fourier transform, and spectral features are extracted. The temporal and spectral features are then input into a convolutional neural network (CNN) and a recurrent neural network (RNN) to automatically extract deep-level features.
[0047] It should be noted that the extraction of time-domain features and spectral features describes the characteristics of voiceprint signals from different perspectives, while convolutional neural networks and recurrent neural networks can automatically extract deep-level features and capture complex patterns of signals. Multi-dimensional feature extraction methods can comprehensively describe the characteristics of voiceprint signals and improve the richness and diversity of features.
[0048] S4. Fuse time-domain features, frequency-domain features, and deep learning features to form a comprehensive feature vector;
[0049] Specifically, the extracted time-domain features, spectral features, and deep learning features are fused to form a comprehensive feature vector, expressed as follows:
[0050] ;
[0051] in, For the comprehensive feature vector, and This is the weighting factor.
[0052] It should be noted that the feature fusion step ensures the comprehensiveness and representativeness of the integrated feature vector, improves the quality of the model's input data, and the integrated feature vector contains information from the time domain, frequency domain, and deep-level features, which can better reflect the operating status of the transformer and enhance the model's detection accuracy and robustness.
[0053] S5. An anomaly detection model is constructed using support vector machines, and an attention mechanism is introduced into the model.
[0054] Specifically, Support Vector Machine (SVM) is selected as the anomaly detection model. An attention mechanism is introduced into the SVM model so that the model can focus on the feature parts that are most helpful for anomaly detection. The comprehensive feature vector is input into the pre-trained support vector machine model, and the result is determined as an anomaly based on the anomaly score.
[0055] It should be noted that Support Vector Machine (SVM) is an effective binary classification model suitable for processing high-dimensional data. After introducing the attention mechanism, the model can focus more on the feature parts that are most helpful for anomaly detection, improving the accuracy and robustness of detection. By judging the anomaly score, the abnormal situation of the transformer can be identified more reliably.
[0056] S6. Input the comprehensive feature vector into the anomaly detection model for detection, obtain the detection results, and record each detection result;
[0057] Specifically, create a structured data table to record the results of each test, and use a CSV file to store the test results. After each test is completed, record the test results in the data table.
[0058] It should be noted that the steps of recording test results ensure the integrity and traceability of the data, facilitating subsequent data analysis and fault diagnosis. The use of structured data tables and CSV files makes data management and querying more convenient, improving the system's practicality and reliability.
[0059] This embodiment also provides a computer device applicable to the transformer acoustic signature anomaly detection method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the transformer acoustic signature anomaly detection method proposed in the above embodiment.
[0060] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.
[0061] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the transformer acoustic signature anomaly detection method as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.
[0062] Experimental data table:
[0063] Table 1: Comparison of Experimental Data for Transformer Acoustic Anomaly Detection
[0064]
[0065] The experimental data table above clearly shows that the method of the present invention has significant advantages over traditional methods in many aspects:
[0066] The method of this invention uses four sensors, while traditional methods 1 and 2 use two and three sensors, respectively. The increased number of sensors can more comprehensively cover the operating status of the transformer and reduce the possibility of missed detections. The sensor layout optimization degree of the method of this invention is 0.90, which is much higher than the 0.75 of traditional method 1 and the 0.80 of traditional method 2. The optimized sensor layout can more effectively capture the acoustic signature signal of the transformer and improve the accuracy of detection.
[0067] The sampling rate of the method of this invention is 20kHz, which is higher than the 10kHz of the traditional method 1 and the 15kHz of the traditional method 2. The higher sampling rate can capture more signal details and improve the signal resolution. The signal-to-noise ratio of the method of this invention after noise removal is 30dB, which is much higher than the 20dB of the traditional method 1 and the 25dB of the traditional method 2. Effective noise removal improves the purity of the signal and reduces the impact of background noise on the detection results.
[0068] The feature extraction time of the method of the present invention is 5 seconds, which is much shorter than the 10 seconds of the traditional method 1 and the 8 seconds of the traditional method 2. The fast feature extraction improves the real-time performance of the system and can respond to the operating status of the transformer more quickly. The anomaly detection accuracy of the method of the present invention is 95%, which is much higher than the 80% of the traditional method 1 and the 85% of the traditional method 2. The higher accuracy means that the method of the present invention can more reliably identify the abnormal conditions of the transformer and reduce the possibility of misjudgment.
[0069] The anomaly detection false negative rate of the method of the present invention is 5%, which is much lower than the 20% of the traditional method 1 and the 15% of the traditional method 2. The lower false negative rate means that the method of the present invention can capture the abnormal conditions of the transformer more comprehensively and reduce the risk of false negatives. The anomaly detection false positive rate of the method of the present invention is 2%, which is much lower than the 10% of the traditional method 1 and the 8% of the traditional method 2. The lower false positive rate means that the method of the present invention can more accurately judge the normal operating status of the transformer and reduce the possibility of false alarms.
[0070] In summary, the multi-sensor layout and high-sensitivity MEMS microphone ensured high resolution and high signal-to-noise ratio of the voiceprint signal. The multi-sensor arrangement can cover different locations of the transformer, thus capturing the transformer's operating status more comprehensively. The extraction of time-domain and spectral features describes the characteristics of the voiceprint signal from different perspectives, while convolutional neural networks and recurrent neural networks can automatically extract deep-level features and capture complex signal patterns. Multi-dimensional feature extraction methods can comprehensively describe the characteristics of the voiceprint signal, improving the richness and diversity of features. The extraction of deep-level features can capture the implicit patterns of the signal, providing rich feature support for subsequent anomaly detection. The feature fusion step ensures the comprehensiveness and representativeness of the comprehensive feature vector, improving the quality of the model's input data. The comprehensive feature vector contains information from the time domain, frequency domain, and deep-level features, which can better reflect the transformer's operating status.
[0071] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method for detecting abnormal acoustic signatures in transformers, characterized in that: include, Multiple high-sensitivity acoustic sensors are used to collect acoustic signals during transformer operation; The acquired voiceprint signals are preprocessed; Extracting the time-domain features, spectral features, and deep-level features of the voiceprint signal; By fusing time-domain features, frequency-domain features, and deep learning features, a comprehensive feature vector is formed. An anomaly detection model is constructed using support vector machines, and an attention mechanism is introduced into the model. The comprehensive feature vector is input into the anomaly detection model for detection, and the detection result is obtained. Record the results of each test.
2. The method for detecting abnormal acoustic signatures in a transformer according to claim 1, characterized in that: The method employs multiple high-sensitivity acoustic signature sensors to collect acoustic signature signals during transformer operation. The specific steps are as follows: Multiple highly sensitive acoustic signature sensors are arranged around the transformer, with at least four sensors distributed at the four corners of the transformer. The expression is as follows: ; in, To optimize the sensor layout, For the number of sensors, For the first The weight of each sensor, For the first Sensor sensitivity; A high-sensitivity MEMS microphone was selected as the voiceprint sensor, with a frequency response range between 20Hz and 20kHz. Use high-sampling-rate data acquisition equipment, with a sampling rate of no less than 20kHz.
3. The method for detecting abnormal acoustic signatures in a transformer according to claim 1, characterized in that: The preprocessing of the acquired voiceprint signals involves the following steps: An adaptive filter is used to remove background noise from the speaker signal, expressed as follows: ; in, This is the denoised voiceprint signal. This is the original voiceprint signal. As a smoothing factor, This is the estimated value of the denoised signal from the previous time step; The collected voiceprint signal is divided into segments of fixed length, with each segment ranging from 1 to 5 seconds in length.
4. The method for detecting abnormal acoustic signatures in a transformer according to claim 1, characterized in that: The specific steps for extracting the time-domain features, spectral features, and deep-level features of the voiceprint signal are as follows: Calculate the temporal characteristics of each voiceprint segment, including mean, variance, zero-crossing rate, and energy; The short-time Fourier transform is used to convert the acoustic signature signal into a frequency domain representation and extract spectral features; By inputting temporal and spectral features into convolutional and recurrent neural networks, deep-level features are automatically extracted. The expression is as follows: ； ； in, For time-domain feature vectors, For spectral feature vectors, It is a convolutional neural network. It is a recurrent neural network. These are deep-seated features.
5. The method for detecting abnormal acoustic signatures in a transformer according to claim 1, characterized in that: The specific steps for fusing time-domain features, frequency-domain features, and deep learning features to form a comprehensive feature vector are as follows: The extracted time-domain features, spectral features, and deep learning features are fused to form a comprehensive feature vector, expressed as: ; in, For the comprehensive feature vector, and This is the weighting factor.
6. The method for detecting abnormal acoustic signatures in a transformer according to claim 1, characterized in that: The anomaly detection model is constructed using support vector machines, and an attention mechanism is introduced into the model. The specific steps are as follows: Support Vector Machine (SVM) is chosen as the anomaly detection model. An attention mechanism is introduced into the SVM model to enable it to focus on the features most helpful for anomaly detection. The expression is as follows: ; ; in, Let be the decision function. The number of support vectors, For Lagrange multipliers, For the first The label of each sample, For kernel function, Bias term, For attention weights, For feature dimension, For the first The weights of each feature, For the first The values of each feature.
7. The method for detecting abnormal acoustic signatures in a transformer according to claim 1, characterized in that: The steps for inputting the comprehensive feature vector into the anomaly detection model for detection and obtaining the detection result are as follows: Combined feature vectors The input is fed into the already trained support vector machine model; The expression for determining whether a result is abnormal based on the anomaly score is: ; in, For the test results, This is the anomaly threshold; if the anomaly score is greater than the threshold... If it is, it is judged as abnormal; otherwise, it is judged as normal.
8. The method for detecting abnormal acoustic signatures in a transformer according to claim 1, characterized in that: The specific steps for recording each test result are as follows: Create a structured data table to record the results of each test; Use CSV files or database tables to store the test results; After each test is completed, the test results are recorded in the data table.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the transformer acoustic signature anomaly detection method according to any one of claims 1 to 8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the transformer acoustic signature anomaly detection method according to any one of claims 1 to 8.