Converter transformer voiceprint monitoring and fault identification method and system
Through unsupervised clustering and MFCC feature extraction combined with recurrent neural network fault diagnosis methods, noise interference and small sample problems in voiceprint monitoring of converter transformers are solved, efficient fault identification and early warning are achieved, and the safety and operation level of the equipment are improved.
Patent Information
- Application Number
- CN202510531919.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-25
- Publication Date
- 2025-09-02
AI Technical Summary
Existing voiceprint monitoring technology is susceptible to environmental noise interference in converter transformers, resulting in signal characteristics distortion. Machine learning models lack generalization capabilities in small sample scenarios, making it difficult to accurately identify faults.
Unsupervised clustering algorithm is used to perform initial screening of voiceprint data, key features are extracted using Mel frequency cepspectral coefficient (MFCC), and fault diagnosis is performed through recurrent neural networks, and combined with the model incremental training mechanism, fault identification of converter transformers is achieved.
It effectively reduces environmental noise interference, improves the accuracy of fault identification and the adaptability of the model, can promptly detect latent defects, and supports the safe operation of the converter transformer.
Smart Images
Figure CN120581028A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power equipment status monitoring and fault diagnosis, and more specifically, to a converter transformer voiceprint monitoring and fault identification method and system. Background Art
[0002] In the construction of a new power system centered around ultra-high voltage (UHV) power grids, the continued expansion of the grid and the growth in the number of devices presents greater challenges in the operation and maintenance of power equipment. Traditional periodic manual inspections are no longer able to meet the stringent reliability requirements of modern power grids. Against this backdrop, the introduction of intelligent fault identification technology has become an inevitable choice for improving inspection efficiency and accuracy. Compared to traditional contact-based offline detection methods such as oil chromatography analysis, voiceprint monitoring technology, with its significant non-invasive and real-time online advantages, provides an innovative solution for assessing the condition of key equipment such as converter transformers. This technology uses acoustic sensors to collect equipment operating sound signals in real time, combined with artificial intelligence algorithms to dynamically assess the health status of the equipment, effectively improving the timeliness and diagnostic accuracy of fault warnings.
[0003] However, current voiceprint monitoring technology still faces multiple technical bottlenecks in engineering applications. First, complex noise interference in converter stations can easily distort acoustic signal characteristics, making traditional machine learning models trained on clean data prone to misjudgment. Second, obtaining fault samples in actual operation and maintenance scenarios is difficult, and the problem of insufficient model generalization ability is particularly prominent in small sample scenarios.
[0004] Therefore, a technology is needed to realize converter transformer voiceprint monitoring and fault identification. Summary of the Invention
[0005] The technical solution of the present invention provides a converter transformer voiceprint monitoring and fault identification method and system to solve the problem of how to perform voiceprint monitoring and fault identification on a converter transformer.
[0006] In order to solve the above problems, the present invention provides a method for converter transformer voiceprint monitoring and fault identification, the method comprising:
[0007] collecting voiceprint data of the converter transformer, and determining whether the collected voiceprint data is abnormal;
[0008] Extract the features of the abnormal voiceprint data using Mel-frequency cepstral coefficients to obtain the key features of the abnormal voiceprint data;
[0009] The key features are input into a trained fault diagnosis model, and the key features are diagnosed based on the fault diagnosis model to determine the fault type of the converter transformer.
[0010] Preferably, the method of extracting features of the initially determined abnormal voiceprint data through Mel-frequency cepstral coefficients to obtain key features of the abnormal voiceprint data includes:
[0011] Dividing the abnormal voiceprint data into frames, and obtaining a spectrum diagram for each frame of abnormal voiceprint data based on discrete Fourier transform;
[0012] Convert the spectrogram into a Mel-spectrogram based on a Mel-filter bank;
[0013] Performing discrete cosine transform on the Mel-frequency spectrum to obtain Mel-frequency cepstral coefficients;
[0014] A preset number of dimensions in the Mel-frequency cepstral coefficients are used as key features of the abnormal voiceprint data.
[0015] Preferably, inputting the key features into a trained fault diagnosis model, diagnosing the key features based on the fault diagnosis model, and determining the fault type of the converter transformer includes:
[0016] The fault diagnosis model is a recurrent neural network, which inputs the Mel-frequency cepstral coefficients into the recurrent neural network stacked in multiple layers, and outputs the fault type of the converter transformer through the recurrent neural network.
[0017] Preferably, the method further comprises training the fault diagnosis model:
[0018] The parameters of the recurrent neural network are adjusted by inputting the Mel-frequency cepstral coefficients.
[0019] Preferably, the method further comprises optimizing the fault diagnosis model:
[0020] evaluating the recurrent neural network based on the accuracy of the fault type identified for the converter transformer;
[0021] Based on the evaluation results, the parameters and structure of the recurrent neural network are adjusted, and the recurrent neural network is updated and iterated until the evaluation results reach a preset expected value.
[0022] According to another aspect of the present invention, a converter transformer voiceprint monitoring and fault identification system is provided, the system comprising:
[0023] a collection unit, configured to collect voiceprint data of the converter transformer and determine whether the collected voiceprint data is abnormal;
[0024] An extraction unit is used to extract features of the abnormal voiceprint data through Mel-frequency cepstral coefficients to obtain key features of the abnormal voiceprint data;
[0025] The result unit is used to input the key features into a trained fault diagnosis model, diagnose the key features based on the fault diagnosis model, and determine the fault type of the converter transformer.
[0026] Preferably, the extraction unit is used to extract features of the initially determined abnormal voiceprint data through Mel-frequency cepstral coefficients to obtain key features of the abnormal voiceprint data, and is also used to:
[0027] Dividing the abnormal voiceprint data into frames, and obtaining a spectrum diagram for each frame of abnormal voiceprint data based on discrete Fourier transform;
[0028] Convert the spectrogram into a Mel-spectrogram based on a Mel-filter bank;
[0029] Performing discrete cosine transform on the Mel-frequency spectrum to obtain Mel-frequency cepstral coefficients;
[0030] A preset number of dimensions in the Mel-frequency cepstral coefficients are used as key features of the abnormal voiceprint data.
[0031] Preferably, the result unit is used to input the key features into a trained fault diagnosis model, diagnose the key features based on the fault diagnosis model, and determine the fault type of the converter transformer, and is also used to:
[0032] The fault diagnosis model is a recurrent neural network, which inputs the Mel-frequency cepstral coefficients into the recurrent neural network stacked in multiple layers, and outputs the fault type of the converter transformer through the recurrent neural network.
[0033] Preferably, the result unit is further used to train the fault diagnosis model:
[0034] The parameters of the recurrent neural network are adjusted by inputting the Mel-frequency cepstral coefficients.
[0035] Preferably, the result unit is further used to optimize the fault diagnosis model:
[0036] evaluating the recurrent neural network based on the accuracy of the fault type identified for the converter transformer;
[0037] Based on the evaluation results, the parameters and structure of the recurrent neural network are adjusted, and the recurrent neural network is updated and iterated until the evaluation results reach a preset expected value.
[0038] The technical solution of the present invention provides a method and system for voiceprint monitoring and fault identification of a converter transformer, wherein the method includes: collecting voiceprint data of the converter transformer, judging whether the collected voiceprint data is abnormal; extracting features of the judged abnormal voiceprint data through Mel-frequency cepstral coefficients to obtain key features of the abnormal voiceprint data; inputting the key features into a trained fault diagnosis model, diagnosing the key features based on the fault diagnosis model, and determining the fault type of the converter transformer. The technical solution of the present invention implements an initial screening of voiceprint data features through an unsupervised clustering algorithm, extracts features of the detected abnormal signals using a Mel-frequency cepstral coefficient (MFCC) algorithm, and then inputs the features into a fault identification model for diagnosis to reduce environmental noise interference. The technical solution of the present invention can provide early warning for timely discovery of early latent defects of converter transformers, and provide new state monitoring technology support for improving the safe operation level of converter transformers. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] A more complete understanding of exemplary embodiments of the present invention may be obtained by referring to the following drawings:
[0040] Figure 1 Flowchart of a converter transformer voiceprint monitoring and fault identification method according to a preferred embodiment of the present invention;
[0041] Figure 2 Flowchart of a voiceprint monitoring and fault identification method according to a preferred embodiment of the present invention;
[0042] Figure 3 1. An example diagram of MFCC multi-dimensional feature extraction steps according to a preferred embodiment of the present invention;
[0043] Figure 4 Flow chart of a small sample fault diagnosis method according to a preferred embodiment of the present invention;
[0044] Figure 5 Schematic diagram of the arrangement of commutation soundprint measurement points according to a preferred embodiment of the present invention;
[0045] Figure 6 A schematic diagram of a wireless communication mode according to a preferred embodiment of the present invention; and
[0046] Figure 7 The figure is a structural diagram of a converter transformer voiceprint monitoring and fault identification system according to a preferred embodiment of the present invention. DETAILED DESCRIPTION
[0047] Exemplary embodiments of the present invention will now be described with reference to the accompanying drawings. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to provide a thorough and complete disclosure of the present invention and to fully convey the scope of the present invention to those skilled in the art. The terminology used in the exemplary embodiments shown in the accompanying drawings is not intended to limit the present invention. In the accompanying drawings, identical elements are denoted by the same reference numerals.
[0048] Unless otherwise specified, the terms used herein (including technical terms) have the meanings commonly understood by those skilled in the art. In addition, it is understood that terms defined in commonly used dictionaries should be understood to have the same meanings as those in the context of the relevant fields, and should not be understood as idealized or overly formal meanings.
[0049] Figure 1 The figure is a flow chart of a converter transformer voiceprint monitoring and fault identification method according to a preferred embodiment of the present invention.
[0050] The existing voiceprint detection technology for converter transformers mainly focuses on fault diagnosis in the absence of environmental noise interference. However, when affected by ambient noise, misjudgment is prone to occur, and most algorithms require training with a large amount of clean data to achieve good recognition results.
[0051] This invention proposes an unsupervised clustering algorithm for initial screening of voiceprint data features. The Mel-Frequency Cepstral Coefficient (MFCC) algorithm is then used to extract features from detected abnormal signals. These features are then fed into a fault recognition model for diagnosis, minimizing environmental noise interference. New normal and faulty samples after diagnosis are automatically injected into the dataset, triggering incremental model training to form a self-evolving fault recognition model, thereby improving fault identification accuracy. This invention can provide early warning for the timely detection of latent defects in converter transformers, offering new state monitoring technology support for improving the safe operation of converter transformers.
[0052] The present invention proposes a converter transformer voiceprint monitoring and fault identification method, such as Figure 2 As shown, during converter transformer operation, voiceprints are collected using a monitoring device. First, an unsupervised clustering algorithm is used to perform an initial screening of voiceprint data features, directly storing those that meet healthy conditions in a voiceprint dataset. A Mel-Frequency Cepstral Coefficient (MFCC) algorithm is used to extract features from detected abnormal signals, which are then fed into a fault recognition model for diagnosis. After expert verification of the diagnostic results, new normal and faulty samples are automatically injected into the dataset, triggering incremental model training. This completes a closed-loop iteration of "data collection - feature extraction - intelligent diagnosis - model optimization," ultimately forming a self-evolving voiceprint monitoring and fault recognition model.
[0053] like Figure 1 As shown, the present invention proposes a converter transformer voiceprint monitoring and fault identification method, the method comprising:
[0054] Step 101: Collect voiceprint data of the converter transformer and determine whether the collected voiceprint data is abnormal;
[0055] The present invention uses acoustic sensors to achieve real-time collection and uploading of converter transformer operating noise, performs preliminary noise reduction and pattern recognition on the data based on unsupervised learning methods, directly stores voiceprint data that meets the health characteristics of the equipment into a standard voiceprint dataset, performs fault diagnosis on suspected fault data, and finally completes manual labeling and correction through an expert system. The labeled fault samples are re-injected into the dataset to achieve dynamic updating and quality optimization of the converter transformer voiceprint dataset.
[0056] Step 102: Extract features of the abnormal voiceprint data using Mel-frequency cepstral coefficients to obtain key features of the abnormal voiceprint data;
[0057] Preferably, the initially determined abnormal voiceprint data is subjected to feature extraction using Mel-frequency cepstral coefficients to obtain key features of the abnormal voiceprint data, including:
[0058] The abnormal voiceprint data is divided into frames, and a spectrum diagram is obtained for each frame of abnormal voiceprint data based on discrete Fourier transform;
[0059] Convert the spectrogram into a Mel-spectrogram based on a Mel-filter bank;
[0060] Perform discrete cosine transform on the Mel frequency spectrum to obtain Mel frequency cepstrum coefficients;
[0061] A preset number of dimensions in the Mel-frequency cepstral coefficients are used as key features of abnormal voiceprint data.
[0062] The present invention presents complex high-dimensional features in the acoustic time domain signal collected, and is interfered by on-site noise, so it is necessary to extract low-dimensional key features for analysis. The Mel-frequency cepstral coefficient (MFCC) method is designed based on the human auditory system to simulate the human ear's perception of sound frequency. First, the audio signal is divided into frames, and a discrete Fourier transform (DFT) is applied to each frame to obtain a spectrum graph. The spectrum graph is then converted into a Mel-spectrogram by applying a Mel filter group, and the discrete cosine transform (DCT) is applied to the Mel-spectrogram to obtain the MFCC coefficients. The first few dimensions of the MFCC coefficients can be used as feature representations. The advantage of this method is that it is highly robust to noise when extracting features, is suitable for the complex environment of converter stations, and has a small amount of calculation, which is conducive to real-time calculation and analysis. Figure 3 shown.
[0063] Step 103: Input the key features into the trained fault diagnosis model, diagnose the key features based on the fault diagnosis model, and determine the fault type of the converter transformer.
[0064] Preferably, the key features are input into a trained fault diagnosis model, and the key features are diagnosed based on the fault diagnosis model to determine the fault type of the converter transformer, including:
[0065] The fault diagnosis model is a recurrent neural network. The Mel-frequency cepstral coefficients are input into a multi-layer stacked recurrent neural network, and the fault type of the converter transformer is output through the recurrent neural network.
[0066] Preferably, the method further includes training a fault diagnosis model:
[0067] Adjust the parameters of the recurrent neural network through the input Mel-frequency cepstral coefficients.
[0068] Preferably, the method further includes optimizing the fault diagnosis model:
[0069] The recurrent neural network is evaluated based on the accuracy of fault type identification of the converter transformer;
[0070] Based on the evaluation results, the parameters and structure of the recurrent neural network are adjusted, and the recurrent neural network is updated and iterated until the evaluation results reach the preset expected values.
[0071] The present invention performs initial feature screening and adopts an unsupervised clustering learning method to effectively reduce the amount of calculation and adapt to edge computing.
[0072] The present invention adopts the MFCC method to extract voiceprint features, which has strong robustness, is suitable for the complex environment of converter stations, and has a small amount of calculation, which is conducive to real-time calculation and analysis.
[0073] The present invention diagnoses small sample faults and proposes a small sample fault diagnosis method based on a gated recurrent unit (GRU), which has strong anti-noise ability and realizes continuous updating of the voiceprint dataset through actual measurement and diagnosis. The model can be continuously updated and iterated to improve recognition accuracy.
[0074] In the installation and communication of the present invention, the installation position of the voiceprint monitoring device and the WAPI connection method are proposed to meet the needs of digital upgrading of the converter station.
[0075] The current voiceprint data of converter transformers has difficulties such as limited labeled samples, many fault types, and scarce fault data. The present invention proposes a small sample fault diagnosis method based on gated recurrent units (GRU), which extracts more voiceprint depth features from a small amount of data, and has strong anti-noise ability, and is suitable for the complex acoustic environment of converter stations. GRU is a special recurrent neural network, which takes the MFCC coefficient matrix obtained by feature extraction as input, enters the stacked GRU network, and outputs the fault type. The training process is to adjust the parameters of the network through the input data so that the model can classify unknown audio data. In the process of model training, GRU combines the needs of actual projects and uses the accuracy of voiceprint recognition as an evaluation indicator. By adjusting model parameters, changing model structure, etc., the model is continuously updated and iterated to achieve good recognition effect. Figure 4 shown.
[0076] The present invention is convenient for installation and communication, and meets the needs of digital upgrade of converter stations. The present invention arranges the soundprint monitoring device on the wall surface of the firewall on the bushing side and the long side of the converter transformer ( Figure 5 The monitoring device is located at two-thirds the height of the converter transformer, along the vertical center axis of the equipment. Voiceprint monitoring data from each converter transformer is wirelessly connected to the WAPI for the corresponding converter transformer bay. Then, via a network switch, it is connected to a server in the information room, where a voiceprint monitoring and fault identification program runs. The server outputs voiceprint monitoring operating status and warning information, which is then connected to the digital system in the converter station control room via the internal network.
[0077] The present invention adopts an unsupervised clustering learning method to perform preliminary noise reduction and pattern recognition on the data, which effectively reduces the amount of calculation, adapts to edge computing, and realizes the dynamic update and quality optimization of the commutation voiceprint dataset.
[0078] The present invention adopts the Mel-Frequency Cepstral Coefficient (MFCC) method to extract voiceprint features, which has strong robustness, is suitable for the complex environment of converter stations, and has low computational complexity, which is conducive to real-time calculation and analysis.
[0079] This paper proposes a small-sample fault diagnosis method based on a gated recurrent unit (GRU). This method extracts more in-depth voiceprint features from a small amount of data, demonstrating strong noise immunity. By continuously updating the voiceprint dataset through field measurements and diagnosis, the model can be continuously updated and iterated to improve recognition accuracy.
[0080] The present invention proposes the installation position of the voiceprint monitoring device and the WAPI connection method to meet the digital upgrade requirements of the converter station. Figure 6 shown.
[0081] Figure 7The figure is a structural diagram of a converter transformer voiceprint monitoring and fault identification system according to a preferred embodiment of the present invention.
[0082] like Figure 7 As shown, the present invention provides a converter transformer voiceprint monitoring and fault identification system, the system comprising:
[0083] The collecting unit 701 is used to collect voiceprint data of the converter transformer and determine whether the collected voiceprint data is abnormal;
[0084] An extraction unit 702 is configured to extract features of the abnormal voiceprint data using Mel-frequency cepstral coefficients to obtain key features of the abnormal voiceprint data;
[0085] Preferably, the extraction unit 702 is used to extract features of the initially determined abnormal voiceprint data through Mel-frequency cepstral coefficients to obtain key features of the abnormal voiceprint data, and is also used to:
[0086] The abnormal voiceprint data is divided into frames, and a spectrum diagram is obtained for each frame of abnormal voiceprint data based on discrete Fourier transform;
[0087] Convert the spectrogram into a Mel-spectrogram based on a Mel-filter bank;
[0088] Perform discrete cosine transform on the Mel frequency spectrum to obtain Mel frequency cepstrum coefficients;
[0089] A preset number of dimensions in the Mel-frequency cepstral coefficients are used as key features of abnormal voiceprint data.
[0090] The result unit 703 is used to input the key features into the trained fault diagnosis model, diagnose the key features based on the fault diagnosis model, and determine the fault type of the converter transformer.
[0091] Preferably, the result unit 703 is used to input the key features into a trained fault diagnosis model, diagnose the key features based on the fault diagnosis model, and determine the fault type of the converter transformer, and is also used to:
[0092] The fault diagnosis model is a recurrent neural network. The Mel-frequency cepstral coefficients are input into a multi-layer stacked recurrent neural network, and the fault type of the converter transformer is output through the recurrent neural network.
[0093] Preferably, the result unit 703 is further used to train the fault diagnosis model:
[0094] Adjust the parameters of the recurrent neural network through the input Mel-frequency cepstral coefficients.
[0095] Preferably, the result unit 703 is further used to optimize the fault diagnosis model:
[0096] The recurrent neural network is evaluated based on the accuracy of fault type identification of the converter transformer;
[0097] Based on the evaluation results, the parameters and structure of the recurrent neural network are adjusted, and the recurrent neural network is updated and iterated until the evaluation results reach the preset expected values.
[0098] A converter transformer voiceprint monitoring and fault identification system according to a preferred embodiment of the present invention corresponds to a converter transformer voiceprint monitoring and fault identification method according to another preferred embodiment of the present invention, and will not be described in detail here.
[0099] Unlike commonly used voiceprint monitoring technologies, the converter transformer voiceprint monitoring and fault identification method proposed in this paper implements a closed-loop iteration of "data acquisition - feature extraction - intelligent diagnosis - model optimization." This method is adaptable to the complex noise environments of converter stations and situations with insufficient data sets. It improves the accuracy and robustness of voiceprint detection technology, reduces computational complexity, and facilitates real-time calculation and analysis. Similar technologies have not been reported in published literature domestically or internationally.
[0100] It will be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented in various computer languages, for example, the object-oriented programming language Java and the interpreted scripting language JavaScript.
[0101] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0102] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.
[0103] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0104] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.
[0105] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.
[0106] The invention has been described above with reference to a few embodiments. However, it is readily apparent to a person skilled in the art that other embodiments than the ones disclosed above are equally within the scope of the invention, as defined by the appended patent claims.
[0107] Generally, all terms used in the claims are to be interpreted according to their ordinary meaning in the technical field, unless explicitly defined otherwise herein. All references to "a / the [means, component, etc.]" are to be interpreted openly as referring to at least one instance of the means, component, etc., unless explicitly stated otherwise. The steps of any method disclosed herein do not necessarily need to be performed in the exact order disclosed, unless explicitly stated otherwise.
Claims
1. A method for converter transformer voiceprint monitoring and fault identification, the method comprising: collecting voiceprint data of the converter transformer, and determining whether the collected voiceprint data is abnormal; Extract the features of the abnormal voiceprint data using Mel-frequency cepstral coefficients to obtain the key features of the abnormal voiceprint data; The key features are input into a trained fault diagnosis model, and the key features are diagnosed based on the fault diagnosis model to determine the fault type of the converter transformer.
2. The method according to claim 1, wherein extracting features of the initially determined abnormal voiceprint data using Mel-frequency cepstral coefficients to obtain key features of the abnormal voiceprint data comprises: Dividing the abnormal voiceprint data into frames, and obtaining a spectrum diagram for each frame of abnormal voiceprint data based on discrete Fourier transform; Convert the spectrogram into a Mel-spectrogram based on a Mel-filter bank; Performing discrete cosine transform on the Mel-frequency spectrum to obtain Mel-frequency cepstral coefficients; A preset number of dimensions in the Mel-frequency cepstral coefficients are used as key features of the abnormal voiceprint data.
3. The method according to claim 2, wherein inputting the key features into a trained fault diagnosis model, diagnosing the key features based on the fault diagnosis model, and determining the fault type of the converter transformer comprises: The fault diagnosis model is a recurrent neural network, which inputs the Mel-frequency cepstral coefficients into the recurrent neural network stacked in multiple layers, and outputs the fault type of the converter transformer through the recurrent neural network.
4. The method according to claim 3, further comprising training the fault diagnosis model: The parameters of the recurrent neural network are adjusted by inputting the Mel-frequency cepstral coefficients.
5. The method according to claim 4, further comprising optimizing the fault diagnosis model: evaluating the recurrent neural network based on the accuracy of the fault type identified for the converter transformer; Based on the evaluation results, the parameters and structure of the recurrent neural network are adjusted, and the recurrent neural network is updated and iterated until the evaluation results reach a preset expected value.
6. A converter transformer voiceprint monitoring and fault identification system, the system comprising: a collection unit, configured to collect voiceprint data of the converter transformer and determine whether the collected voiceprint data is abnormal; An extraction unit is used to extract features of the abnormal voiceprint data through Mel-frequency cepstral coefficients to obtain key features of the abnormal voiceprint data; The result unit is used to input the key features into a trained fault diagnosis model, diagnose the key features based on the fault diagnosis model, and determine the fault type of the converter transformer.
7. The system according to claim 6, wherein the extraction unit is configured to extract features of the initially determined abnormal voiceprint data using Mel-frequency cepstral coefficients to obtain key features of the abnormal voiceprint data, and is further configured to: Dividing the abnormal voiceprint data into frames, and obtaining a spectrum diagram for each frame of abnormal voiceprint data based on discrete Fourier transform; Convert the spectrogram into a Mel-spectrogram based on a Mel-filter bank; Performing discrete cosine transform on the Mel-frequency spectrum to obtain Mel-frequency cepstral coefficients; A preset number of dimensions in the Mel-frequency cepstral coefficients are used as key features of the abnormal voiceprint data.
8. The system according to claim 7, wherein the result unit is configured to input the key features into a trained fault diagnosis model, diagnose the key features based on the fault diagnosis model, and determine the fault type of the converter transformer, and is further configured to: The fault diagnosis model is a recurrent neural network, which inputs the Mel-frequency cepstral coefficients into the recurrent neural network stacked in multiple layers, and outputs the fault type of the converter transformer through the recurrent neural network.
9. The system according to claim 8, wherein the result unit is further configured to train the fault diagnosis model: The parameters of the recurrent neural network are adjusted by inputting the Mel-frequency cepstral coefficients.
10. The system according to claim 9, wherein the result unit is further configured to optimize the fault diagnosis model: evaluating the recurrent neural network based on the accuracy of the fault type identified for the converter transformer; Based on the evaluation results, the parameters and structure of the recurrent neural network are adjusted, and the recurrent neural network is updated and iterated until the evaluation results reach a preset expected value.