Transformer voiceprint fault recognition system and method
By collecting transformer voiceprint samples and performing feature extraction and deep learning algorithm recognition, the problem of low accuracy of transformer fault diagnosis in the prior art is solved, and the diagnosis efficiency and accuracy are improved.
Patent Information
- Application Number
- CN202311328455.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-13
- Publication Date
- 2025-04-15
AI Technical Summary
The accuracy of existing transformer fault diagnosis methods is low, especially when the sample data is insufficient.
By collecting the transformer soundprint samples, calculate the change in the voiceprint component. If the set threshold is exceeded, the time domain features, frequency domain features and cepspectral parameters will be extracted, and preliminary fault judgment will be made based on the frequency domain parameter information. If it is not recognized, a deep learning algorithm will be used for identification.
It improves the accuracy of transformer fault diagnosis, saves time to analyze samples without abnormal soundprints, and improves diagnostic efficiency.
Smart Images
Figure CN119830060A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a transformer voiceprint fault recognition system and method, belonging to the technical field of transformer fault diagnosis. Background Art
[0002] As the core power equipment of the power system, the safe and stable operation of the power transformer directly affects the reliable transmission and flexible distribution of the entire power system. With the rapid development of the economy and the growth of electricity demand, the voltage level of the power system has been continuously improved. Modern large transformers have the characteristics of large rated capacity, high price, and diverse structure. Once a power transformer fails, the impact range is wide, the power outage load is large, and the consequences will be unpredictable. Therefore, in order to eliminate hidden dangers, prevent accidents, and improve power supply reliability, various measures are adopted to monitor the operating status of power transformers, identify existing faults and hidden dangers, and ensure the safe and stable operation of power transformers.
[0003] At present, transformer fault diagnosis is mainly carried out through voiceprint diagnosis. Voiceprint diagnosis methods are mainly divided into machine learning methods and eigenvalue diagnosis methods. Machine learning methods can train more powerful models from a large number of sample data sets for fault diagnosis, but machine learning methods require a large number of transformer voiceprint samples, especially fault samples and stable and reliable models to achieve excellent results. When the sample data is small, the diagnosis is inaccurate; the eigenvalue diagnosis method mainly performs mechanism analysis from the vibration source of the transformer voiceprint signal to perform fault diagnosis, but there is no unified analysis and threshold setting for common transformer faults, and the warning thresholds for various characteristic indicators when applied on site are also relatively vague, resulting in low accuracy of transformer fault diagnosis. Summary of the invention
[0004] The purpose of the present invention is to provide a transformer voiceprint fault identification system and method to solve the problem of low accuracy in the current fault diagnosis process.
[0005] In order to solve the above technical problems, the present invention provides a transformer voiceprint fault identification method, which comprises the following steps:
[0006] 1) Collect transformer voiceprint samples and calculate the change of voiceprint components through the voiceprint samples;
[0007] 2) If the change in the voiceprint component exceeds the set threshold, the time domain features, frequency domain features and cepstrum parameters of the voiceprint sample are extracted;
[0008] 3) Perform fault judgment on the voiceprint sample based on the extracted frequency domain parameter information and identify the fault type;
[0009] 4) When the fault type cannot be identified based on the frequency domain parameter information, a deep learning algorithm is used for identification based on the time domain characteristics, frequency domain characteristics and cepstrum parameter information of the voiceprint sample.
[0010] The present invention first collects voiceprint samples of the transformer, makes judgments based on the changes in voiceprint components, extracts features from voiceprint samples when the changes in voiceprint components exceed a set threshold, and identifies transformer faults by frequency domain analysis and deep learning algorithms on voiceprint samples, thereby saving time for analyzing voiceprint samples that are not abnormal and improving diagnostic efficiency. The present invention combines multiple transformer fault diagnosis methods, conducts comprehensive analysis and identification of transformer voiceprints, and improves the accuracy of transformer fault diagnosis.
[0011] Furthermore, the change in the voiceprint component is:
[0012]
[0013] where f max is the upper limit frequency of the calculation, A1(n×50) is the n×50 Hz frequency component amplitude of the normal standard voiceprint sample signal, and A2(n×50) is the n×50 Hz frequency component amplitude of the voiceprint sample signal to be tested.
[0014] Furthermore, the time domain features include zero-crossing rate, short-time energy spectrum and power spectrum, which are obtained by calculating the preprocessed voiceprint samples.
[0015] Furthermore, the frequency domain features include main frequency, vibration entropy, fundamental frequency proportion, 50Hz odd-even multiple ratio and high-low frequency ratio, which are obtained by Fourier transforming the preprocessed voiceprint samples.
[0016] Furthermore, the cepstral parameters are Mel cepstral coefficients, which are obtained by performing triangular filtering, log energy, and DFT cosine transform on the preprocessed voiceprint samples.
[0017] Furthermore, in step 2), if the change in the voiceprint component exceeds a set threshold, an alarm message is issued to give an alarm.
[0018] Furthermore, in step 2), if the change in the voiceprint component does not exceed the set threshold, the corresponding transformer is not faulty.
[0019] Furthermore, the preprocessing includes normalization, pre-emphasis, frame division and windowing.
[0020] Furthermore, the deep learning algorithm includes convolutional neural network CNN, LSTM or XGBoost.
[0021] A transformer voiceprint fault identification system includes a memory and a processor, wherein the memory is used to receive and store collected voiceprint samples and transmit the voiceprint samples to the processor; the processor is used to calculate the change amount of voiceprint components through the voiceprint samples; and when the change amount of the voiceprint components exceeds a set threshold, the voiceprint samples are feature extracted, and the voiceprint samples are judged according to the extracted frequency domain parameter information to identify the fault type. When the fault type cannot be identified according to the frequency domain parameter information, a deep learning algorithm is used for identification according to the time domain characteristics, frequency domain characteristics and cepstrum parameter information of the voiceprint samples.
[0022] The present invention first collects voiceprint samples of the transformer, makes judgments based on the changes in voiceprint components, extracts features from voiceprint samples when the changes in voiceprint components exceed a set threshold, and identifies transformer faults by frequency domain analysis and deep learning algorithms on voiceprint samples, thereby saving time for analyzing voiceprint samples that are not abnormal and improving diagnostic efficiency. The present invention combines multiple transformer fault diagnosis methods, conducts comprehensive analysis and identification of transformer voiceprints, and improves the accuracy of transformer fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1 is a flow chart of the transformer voiceprint fault identification method of the present invention;
[0024] Figure 2 4 is a flow chart of voiceprint feature extraction in an embodiment of the present invention. DETAILED DESCRIPTION
[0025] The specific implementation of the present invention will be further described below in conjunction with the accompanying drawings.
[0026] Embodiment of transformer voiceprint fault identification method
[0027] The present invention first collects transformer voiceprint samples through a microphone sensor, and calculates the change in voiceprint components through the voiceprint samples; if the change in voiceprint components exceeds the set threshold, an alarm message is issued and the voiceprint samples whose change in voiceprint components exceeds the set threshold are marked, and features are extracted from the voiceprint samples; the marked voiceprint samples are judged based on the frequency domain parameter information extracted from the voiceprint sample features to identify the fault type; when the fault type cannot be identified based on the frequency domain parameter information, a deep learning algorithm is used for identification. The implementation process of this method is as follows: Figure 1 As shown, the following is explained with reference to specific examples.
[0028] 1. Collect transformer voiceprint samples and calculate the change in voiceprint components through the voiceprint samples.
[0029] The voiceprint component change is calculated through the voiceprint sample. The calculation formula of the voiceprint component change D is:
[0030]
[0031] Where fmax is the calculated upper limit frequency, A1(n×50) is the n×50Hz frequency component amplitude of the normal standard voiceprint sample signal, and A2(n×50) is the n×50Hz frequency component amplitude of the voiceprint sample signal to be tested.
[0032] 2. Analyze the component changes of voiceprint samples.
[0033] When the voiceprint component change exceeds the set threshold, an alarm message is issued and the voiceprint sample whose voiceprint component change exceeds the set threshold is marked, and the time domain features, frequency domain features and cepstrum parameters of the voiceprint sample are extracted. In this embodiment, the voiceprint component change threshold is set to 7%, which can be adjusted.
[0034] The voiceprint samples are subjected to feature extraction, and the time domain features, frequency domain features, and Mel-frequency cepstral coefficients (MFCC) and other cepstral parameters of the voiceprint samples are extracted. The transformer voiceprint feature extraction process is as follows: Figure 2 As shown in the figure: First, the input sample is preprocessed, including normalization, pre-emphasis, framing and windowing, and the time domain characteristics of the voiceprint sample are calculated, including zero-crossing rate, short-time energy spectrum and power spectrum. After preprocessing, a short-time Fourier transform is performed to obtain the frequency domain characteristics of the voiceprint sample. According to the previous analysis and summary, the main frequency, vibration entropy, fundamental frequency ratio, 50Hz odd-even submultiple ratio and high-low frequency ratio that can reflect the frequency domain characteristics of the voiceprint sample are calculated. Then, through the triangular filter group, the log logarithmic energy is calculated, and finally the Mel cepstral coefficient MFCC of the voiceprint sample is obtained through the DFT cosine transform. The time domain characteristics, frequency domain characteristics and MFCC after dimension reduction of the voiceprint sample are combined to obtain the comprehensive characteristics of the voiceprint sample.
[0035] Pre-emphasis is a signal processing method that compensates for the high-frequency components of the input signal at the transmitting end. Pre-emphasis is to enhance the high-frequency components of the signal at the beginning of the transmission line to compensate for the excessive attenuation of the high-frequency components during the transmission process.
[0036] 3. Judge the marked voiceprint samples based on the frequency domain parameter information extracted from the voiceprint sample features to identify the fault type.
[0037] According to the frequency domain parameter information extracted from the voiceprint samples, the voiceprint samples whose marked component changes exceed the threshold are judged and analyzed. The frequency domain parameters mainly include main frequency, vibration entropy, fundamental frequency ratio, 50Hz odd-even multiple frequency ratio and high-low frequency ratio, etc., to identify common fault types with obvious characteristics. By setting the threshold of the main frequency domain parameters, when the frequency domain parameters of the voiceprint sample exceed the threshold, it is a fault sample. According to the changes in different frequency domain parameters caused by common fault types, the specific fault type of the sample is judged. For example, a DC bias fault will cause the transformer voiceprint main frequency to increase, the 50Hz odd multiple frequency to increase, the high-low frequency ratio to increase, and the fundamental frequency ratio to decrease; heavy overload causes the overall amplitude capacity of the transformer voiceprint fundamental frequency to increase, and the high-low frequency ratio to decrease.
[0038] 4. When the fault type cannot be identified based on the frequency domain parameter information, a deep learning algorithm is used for identification.
[0039] For fault types that are difficult to distinguish, such as short-circuit shock and loose transformer components, it is necessary to further use algorithm models for analysis. After extracting the characteristic parameters of voiceprint samples of known fault types and reducing the dimension, the deep learning algorithm is used to train the model. The characteristic parameters extracted from the fault voiceprint samples that need to be judged are input into the trained model for fault type identification. The deep learning algorithm can use convolutional neural networks CNN, LSTM and XGBoost, etc.
[0040] Embodiment of transformer voiceprint fault identification system
[0041] The present invention includes a memory and a processor. The memory is used to receive and store collected voiceprint samples, and transmit the voiceprint samples to the processor; the processor is used to calculate the change in voiceprint components through the voiceprint samples; and when the change in voiceprint components exceeds a set threshold, feature extraction is performed on the voiceprint samples, and the marked voiceprint samples are judged according to the extracted frequency domain parameter information to identify the fault type. When the fault type cannot be identified according to the frequency domain parameter information, a deep learning algorithm is used for identification according to the time domain characteristics, frequency domain characteristics and cepstrum parameter information of the voiceprint samples.
Claims
1. A transformer voiceprint fault identification method, characterized in that: The method comprises the following steps: 1) Collect transformer voiceprint samples and calculate the change of voiceprint components through the voiceprint samples; 2) If the change in the voiceprint component exceeds the set threshold, the time domain features, frequency domain features and cepstrum parameters of the voiceprint sample are extracted; 3) Perform fault judgment on the voiceprint sample based on the extracted frequency domain parameter information and identify the fault type; 4) When the fault type cannot be identified based on the frequency domain parameter information, a deep learning algorithm is used for identification based on the time domain characteristics, frequency domain characteristics and cepstrum parameter information of the voiceprint sample.
2. The transformer voiceprint fault identification method according to claim 1, characterized in that: The change in the voiceprint component is: where f max is the upper limit frequency of the calculation, A1(n×50) is the n×50 Hz frequency component amplitude of the normal standard voiceprint sample signal, and A2(n×50) is the n×50 Hz frequency component amplitude of the voiceprint sample signal to be tested.
3. The transformer voiceprint fault identification method according to claim 1, characterized in that: The time domain features include zero-crossing rate, short-time energy spectrum and power spectrum, which are obtained by calculating the preprocessed voiceprint samples.
4. The transformer voiceprint fault identification method according to claim 3 is characterized in that: The frequency domain features include main frequency, vibration entropy, fundamental frequency proportion, 50Hz odd-even multiple frequency ratio and high-low frequency ratio, which are obtained by Fourier transforming the preprocessed voiceprint samples.
5. The transformer voiceprint fault identification method according to claim 1, characterized in that: The cepstrum parameters are Mel cepstrum coefficients, which are obtained by performing triangular filtering, log energy, and DFT cosine transformation on the preprocessed voiceprint samples.
6. The transformer voiceprint fault identification method according to claim 1, characterized in that: In the step 2), if the change in the voiceprint component exceeds a set threshold, an alarm message is issued to give an alarm.
7. The transformer voiceprint fault identification method according to claim 1, characterized in that: In the step 2), if the change in the voiceprint component does not exceed the set threshold, the corresponding transformer is not faulty.
8. The transformer voiceprint fault identification method according to claim 3, characterized in that: The preprocessing includes normalization, pre-emphasis, frame division and windowing.
9. The transformer voiceprint fault identification method according to claim 1, characterized in that: The deep learning algorithm includes convolutional neural network CNN, LSTM or XGBoost.
10. A transformer voiceprint fault identification system, characterized in that: It includes a memory and a processor. The memory is used to receive and store the collected voiceprint samples and transmit the voiceprint samples to the processor. The processor is used to perform diagnosis using the transformer voiceprint fault identification method described in 1-9.