Transformer voiceprint detection method and device, electronic equipment and storage medium
By collecting transformer acoustic fingerprint information, performing feature extraction and model training, and utilizing clustering and independent component analysis algorithms, the problem of accurate identification of transformer faults in complex environments was solved, achieving efficient and accurate fault type identification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HANGZHOU SHENGGUAN TECH CO LTD
- Filing Date
- 2023-03-29
- Publication Date
- 2026-07-28
AI Technical Summary
Existing transformer fault detection methods are ineffective in complex environments, cannot effectively identify faults such as loose internal components, and fail to consider the influence of various factors such as noise.
Acoustic fingerprint information of transformers under ideal operating conditions is collected, and Mel-frequency cepstral coefficients are obtained through feature extraction to train an initial detection model. Acoustic fingerprint information is collected under complex environments for feature extraction and type labeling. Clustering algorithm and fast independent component analysis algorithm are used to train the final detection model to achieve fault type identification.
It enables efficient and accurate identification of transformer fault types in complex environments, improving the accuracy and reliability of detection.
Smart Images
Figure CN116364108B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of transformer fault detection, and in particular to a transformer acoustic signature detection method and device, electronic equipment, and storage medium. Background Technology
[0002] During long-term operation, transformers will experience various faults as their service life increases. Currently, methods such as dissolved gas analysis, infrared thermography, and partial discharge monitoring are generally used to detect transformer faults.
[0003] On the other hand, the above methods are not effective in detecting the cause of faults such as loose screws in internal transformer components. Since the sound of a transformer during operation contains a wealth of information about its operating status, and the sound emitted during normal operation has certain regularities, abnormal sounds will be produced when there is a mechanical fault inside the transformer. Therefore, abnormal sounds emitted by the transformer can be detected, and transformer fault diagnosis can be performed by observing changes in the transformer's acoustic signature.
[0004] However, most existing detection methods are based on idealized data models and do not take into account the influence of various factors such as noise in complex environments. When using the above-mentioned idealized models to detect transformer acoustic fingerprint information, the detection effect will continuously decline, thus making it impossible to accurately identify transformer faults. Summary of the Invention
[0005] The purpose of this application is to solve the technical problems mentioned in the background art and to provide a transformer acoustic fingerprint detection method and device, electronic device, and storage medium.
[0006] In a first aspect, embodiments of this application provide a transformer acoustic signature detection method, comprising: collecting first acoustic signature information of a transformer, wherein the first acoustic signature information is the sound information of the transformer under ideal operating conditions; extracting features from the first acoustic signature information to obtain first Mel-Cepstral Coefficients; inputting the first Mel-Cepstral Coefficients into a preset neural network for training to obtain an initial detection model; collecting second acoustic signature information of a transformer, wherein the second acoustic signature information is the sound information of the transformer under complex environments; extracting features from the second acoustic signature information to obtain second Mel-Cepstral Coefficients; labeling the types of the second Mel-Cepstral Coefficients using a clustering algorithm and a fast independent component analysis algorithm; training the initial detection model using the labeled second Mel-Cepstral Coefficients to obtain a final detection model; and detecting the acoustic signature information of the transformer to be tested according to the final detection model to achieve identification of transformer fault types.
[0007] Optionally, the step of extracting features from the first voiceprint information to obtain the first Mel-frequency cepstral coefficients specifically includes: performing high-pass filtering on the first voiceprint information to filter out low-frequency components; performing frame segmentation on the first voiceprint information after high-pass filtering; performing windowing on the framed first voiceprint information; performing time-frequency transformation on the windowed first voiceprint information; extracting the frequency response of the time-frequency transformed first voiceprint information; and performing logarithmic and cosine transformations on the frequency response to obtain the first Mel-frequency cepstral coefficients.
[0008] Optionally, the step of labeling the type of the second Mel-Cepstral coefficient using clustering algorithms and fast independent component analysis algorithms specifically includes: using the first Mel-Cepstral coefficient as the cluster center, performing clustering calculations on the second Mel-Cepstral coefficient; labeling the second Mel-Cepstral coefficients whose distance from the cluster center is less than a first distance threshold as the type of the corresponding cluster center; and performing fast independent component analysis on the second Mel-Cepstral coefficients whose distance from the cluster center is greater than a second distance threshold, and labeling them as the corresponding types according to the analysis results.
[0009] Optionally, the step of clustering the second Mel-Cepstral Coefficient with the first Mel-Cepstral Coefficient as the cluster center specifically includes using the formula: d = 1 - IOU; Clustering calculations are performed, where x1 represents the first dimension feature in the first Mel-Cepstral coefficients, x'1 represents the first dimension feature in the second Mel-Cepstral coefficients, x2 represents the second dimension feature in the first Mel-Cepstral coefficients, and x'2 represents the second dimension feature in the second Mel-Cepstral coefficients. n x' represents the nth dimension feature in the first Mel-Cepstral coefficients. n d represents the nth dimension feature in the second Mel-Cepstral coefficients, K represents a constant, and d represents the distance to the cluster center.
[0010] Optionally, the first distance threshold is 0.3 and the second distance threshold is 0.7.
[0011] Optionally, before labeling the type of the second Mel-Cepstral Coefficient using clustering algorithms and fast independent component analysis algorithms, the method further includes: scoring the second Mel-Cepstral Coefficient by calculating the error between the true value and the predicted value of the second Mel-Cepstral Coefficient, wherein the second Mel-Cepstral Coefficient with a score lower than the first score is used for type labeling under subsequent clustering algorithms and fast independent component analysis algorithms, and the second Mel-Cepstral Coefficient with a score higher than the second score is used to train the initial detection model.
[0012] Secondly, embodiments of this application provide a transformer acoustic signature detection device, comprising: a first acquisition module configured to acquire first acoustic signature information of a transformer, wherein the first acoustic signature information is the sound information of the transformer under ideal operating conditions; a first feature extraction module configured to extract features from the first acoustic signature information to obtain first Mel-Cepstral Coefficients; a first training module configured to input the first Mel-Cepstral Coefficients into a preset neural network for training to obtain an initial detection model; a second acquisition module configured to acquire second acoustic signature information of a transformer, wherein the second acoustic signature information is the sound information of the transformer under complex environments; a second feature extraction module configured to extract features from the second acoustic signature information to obtain second Mel-Cepstral Coefficients; a type labeling module configured to label the type of the second Mel-Cepstral Coefficients using a clustering algorithm and a fast independent component analysis algorithm; a second training module configured to train the initial detection model using the labeled second Mel-Cepstral Coefficients to obtain a final detection model; and a detection module configured to detect the acoustic signature information of the transformer to be tested according to the final detection model to achieve identification of transformer fault types.
[0013] Optionally, the type labeling module includes: a center setting submodule, configured to perform clustering calculations on the second Mel-Cepstral Coefficients using the first Mel-Cepstral Coefficient as the cluster center; a first labeling submodule, configured to label the second Mel-Cepstral Coefficients whose distance from the cluster center is less than a first distance threshold as the type of the corresponding cluster center; and a second labeling submodule, configured to perform rapid independent component analysis on the second Mel-Cepstral Coefficients whose distance from the cluster center is greater than a second distance threshold, and label them as the corresponding types according to the analysis results.
[0014] Thirdly, embodiments of this application provide an electronic device, including a memory and a processor. The memory stores computer-readable instructions, which, when executed by the processor, cause the processor to perform the aforementioned transformer acoustic signature detection method.
[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when run on a computer, causes the computer to execute the aforementioned transformer acoustic signature detection method.
[0016] The beneficial effects of this application are:
[0017] The transformer acoustic signature detection method provided in this application collects first acoustic signature information of the transformer under ideal operating conditions, and trains a preset neural network to obtain an initial detection model based on the first Mel-Cepstral Coefficients obtained by feature extraction of the first acoustic signature information. Then, it collects second acoustic signature information of the transformer under complex environments, and performs corresponding type labeling on the second Mel-Cepstral Coefficients extracted from the second acoustic signature information. The type-labeled second Mel-Cepstral Coefficients are then input into the initial detection model to train a final detection model. The final detection model is used to detect the acoustic signature information to be tested, and finally, it can efficiently and accurately identify the fault type of the transformer. Attached Figure Description
[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly described below.
[0019] Figure 1 This is a schematic flowchart of a transformer acoustic fingerprint detection method according to an embodiment of this application;
[0020] Figure 2 This is a schematic diagram of the process for extracting the first Mel-frequency cepstral coefficients according to an embodiment of this application;
[0021] Figure 3 This is a schematic diagram illustrating the process of labeling the type of the second Mel-frequency cepstral coefficients according to an embodiment of this application;
[0022] Figure 4 This is a block diagram of a transformer acoustic signature detection device according to an embodiment of this application;
[0023] Figure 5 This is a block diagram of a type labeling module in one embodiment of this application;
[0024] Figure 6 This is a device block diagram of an electronic device provided in one embodiment of this application. Detailed Implementation
[0025] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0026] This application first provides a transformer acoustic signature detection method for detecting acoustic signature information to achieve efficient and accurate identification of transformer fault types.
[0027] Please refer to Figure 1 This is a flowchart illustrating a transformer acoustic signature detection method provided in an embodiment of this application. The method includes steps S110 to S180.
[0028] Step S110: Collect the first acoustic fingerprint information of the transformer, which is the sound information of the transformer under ideal operating conditions.
[0029] Under ideal operating conditions, the frequency of transformer sound is mainly distributed in the range of 0 to 2000 Hz, and is often an even harmonic of 50 Hz.
[0030] In this step, the sound information of the transformer under ideal operating conditions is collected by a sensor and used as the first voiceprint information.
[0031] According to the Nyquist sampling theorem, an audio sample is taken from the transformer under ideal conditions using a first sampling frequency that is more than twice the highest frequency of the signal. Since the frequency of the transformer sound is mainly distributed in the range of 0 to 2000 Hz under ideal operating conditions, the first sampling frequency is at least 4000 Hz, thus ensuring that the first voiceprint information collected does not lose any information.
[0032] For example, a sensor with a certain sensitivity can be used to sample at a first sampling frequency. Specifically, sound information can be collected by fixing the total sampling duration or the total number of samples to obtain the first acoustic signature information of the transformer. Furthermore, after collecting the first acoustic signature information, processing can be performed on it, including filtering, noise reduction, and amplification.
[0033] Step S120: Extract features from the first voiceprint information to obtain the first Mel-frequency cepstral coefficients.
[0034] The first voiceprint information contains different feature information. By extracting features from the first voiceprint information, the first Mel-Cepstral Coefficient is obtained, which is used to characterize the first voiceprint information for training the subsequent detection model.
[0035] Please refer to Figure 2 The diagram below is a flowchart illustrating the extraction of the first Mel-Cepstral Coefficients according to an embodiment of this application. Specifically, the extraction of features from the first voiceprint information to obtain the first Mel-Cepstral Coefficients includes steps S1210 to S1260.
[0036] Step S1210: Perform high-pass filtering on the first voiceprint information to filter out the low-frequency components in the first voiceprint information.
[0037] For example, the first voiceprint information collected can be input into the system function as: H(Z) = 1 - μz -1 In the filter, where the filter is a high-pass filter, and μ is typically set to 0.97. By performing high-pass filtering, the low-frequency components in the first voiceprint information are filtered out, thereby enhancing the high-frequency components in the first voiceprint information and making the spectrum of the first voiceprint information more stable.
[0038] It is understood that filters with different parameters can be used to perform high-pass filtering on the first voiceprint information. The above is only an example and is not intended to limit this application.
[0039] Step S1220: Perform frame segmentation on the first voiceprint information that has undergone high-pass filtering.
[0040] In this step, N sampling points are used as one frame to segment the first voiceprint information after high-pass filtering for subsequent data processing. Furthermore, partial overlap between adjacent frames can be used to avoid excessive differences between them.
[0041] Step S1230: Window the first voiceprint information after framing.
[0042] To perform time-frequency transformation on the first voiceprint information, it is necessary to first perform windowing processing on the framed first voiceprint information. Among them, the window types in the windowing processing are usually rectangular windows, Gaussian windows, and Hamming windows.
[0043] In one embodiment, the first audioprint information after framing is combined with a Hamming window function: Multiply the results to obtain the first voiceprint information after applying the Hamming window. It is understood that, in addition to using the Hamming window function, other types of window functions can be used for windowing processing; the above are merely examples and are not intended to limit this application.
[0044] Step S1240: Perform time-frequency transformation on the windowed first voiceprint information.
[0045] In this step, the windowed first voiceprint information is transformed from the time domain to the frequency domain using a Fast Fourier Transform. The spectrum of each frame of the first voiceprint information after time-frequency transformation can be expressed as:
[0046]
[0047] Step S1250: Extract the frequency response of the first voiceprint information after time-frequency transformation.
[0048] In this step, a Mel-scale triangular filter is used to extract the frequency response of the first voiceprint information spectrum, where the Mel frequency can be expressed as: The frequency response of the extracted first voiceprint information is:
[0049]
[0050] Step S1260: Perform logarithmic operation and cosine transform on the frequency response to obtain the first Mel-frequency cepstral coefficients.
[0051] In this step, the frequency response function H of the first voiceprint information is first evaluated. m (k) Perform logarithmic operations: Then substitute the result of the logarithmic operation into the discrete cosine transform: The first Mel-Cepstral coefficients C(n) are obtained. Where L is the order of the first Mel-Cepstral coefficients and M is the number of filters.
[0052] Through the above steps S1210 to S1260, the first Mel-frequency cepstral coefficients are extracted from the first acoustic signature information of the transformer.
[0053] Step S130: Input the first Mel-frequency cepstral coefficients into a preset neural network for training to obtain an initial detection model.
[0054] In this step, initial weights, thresholds, hidden layers, and the number of nodes in each layer are set to establish a preset neural network. The first Mel-Cepstral coefficients are input into the preset neural network for training. By calculating the error between the output value and the predicted value, the preset neural network that does not meet the minimum error condition is adaptively adjusted by modifying the number of hidden layers and nodes until the minimum error condition is met, thus obtaining the initial detection model.
[0055] In this embodiment, a back-propagation artificial neural network (BP artificial neural network) is used as the preset neural network. The number of nodes in the input layer, output layer, and hidden layer of the preset neural network satisfies the following formula: Where l is the number of hidden layer nodes, m is the number of output layer nodes, n is the number of input layer nodes, and a is a constant between 0 and 10.
[0056] Accordingly, the training loss function of the preset neural network is the mean squared error function, which can be expressed as:
[0057]
[0058] In the formula, yi, f xi Let be the true value of the i-th sample and its corresponding predicted value, respectively, and n be the number of samples.
[0059] In one embodiment, a portion of the first Mel-Cepstral coefficients can be used as a first training set, and another portion of the first Mel-Cepstral coefficients can be used as a first test set. The first training set is used to input into a preset neural network for training to obtain an intermediate initial detection model, and the first test set is used to input into the aforementioned first initial detection model for training to obtain a final initial detection model, i.e., the initial detection model.
[0060] Step S140: Collect the second acoustic fingerprint information of the transformer, which is the sound information of the transformer in a complex environment.
[0061] In complex environments, the sound information of a transformer is affected by a variety of factors, including transformer load, ambient temperature, weather conditions, transformer model, and noise.
[0062] Since the frequency of a normally functioning transformer is mainly distributed in the range of 0–2000Hz, while the frequency of a mechanical fault is distributed in the range of 0–4000Hz, there is a certain overlap in frequency ranges between the operating sound patterns of a transformer and the sound patterns of a fault. In addition, there are various irrelevant frequencies of noise in complex environments, such as the sound of human speech in the range of 250–5000Hz.
[0063] In this step, sensors are used to collect the sound information of the transformer under complex environments, which serves as the second acoustic signature information. Sound information above 4000Hz is generally considered to be other irrelevant noise and is unrelated to the fault acoustic signature, so irrelevant noise above 4000Hz can be filtered out first.
[0064] According to the Nyquist sampling theorem, a second sampling frequency greater than twice the highest frequency of the signal is used to sample the audio of the transformer in a complex environment. The second sampling frequency is at least 8000Hz, so as to ensure that the collected second voiceprint information does not lose any information.
[0065] Step S150: Extract features from the second voiceprint information to obtain the second Mel-frequency cepstral coefficients.
[0066] The second voiceprint information contains different feature information. By extracting features from the second voiceprint information, the second Mel-Cepstral Coefficient is obtained, which is used to characterize the second voiceprint information for training the subsequent detection model.
[0067] The specific steps for extracting features from the second voiceprint information to obtain the second Mel-Cepstral Coefficient are similar to those for extracting the first Mel-Cepstral Coefficient. Please refer to the descriptions in steps S1210 to S1260 for details, which will not be repeated here.
[0068] In one embodiment, the error between the true value and the predicted value of the second Mel-Cepstral Coefficient can be calculated using the mean square error function of a preset neural network to score the second Mel-Cepstral Coefficient.
[0069] Among them, the second Mel-Cepstral coefficients with scores lower than the first score are used for type labeling in subsequent clustering algorithms and fast independent component analysis algorithms, while the second Mel-Cepstral coefficients with scores higher than the second score are used to train the initial detection model.
[0070] For example, the first score is 0.3 and the second score is 0.7. It is understood that the first and second scores can be set to other values; the above are merely examples and are not intended to limit this application.
[0071] Step S160: Use clustering algorithm and fast independent component analysis algorithm to label the type of the second Mel-frequency cepstral coefficients.
[0072] If the second Mel-Cepstral coefficients are directly input into the initial detection model for training, the presence of other related noise terms in the second Mel-Cepstral coefficients can easily lead to model recognition errors. This embodiment uses clustering algorithms and fast independent component analysis algorithms to label the types of the second Mel-Cepstral coefficients, thereby achieving noise reduction and separation.
[0073] Please refer to Figure 3 This is a flowchart illustrating the labeling of the type of the second Mel-frequency cepstral coefficients according to an embodiment of this application, specifically including steps S1610 to S1630.
[0074] Step S1610: Using the first Mel-Cepstral coefficient as the cluster center, perform cluster calculation on the second Mel-Cepstral coefficient.
[0075] In one embodiment, the following formula is used:
[0076] d = 1 - IOU;
[0077]
[0078] Perform clustering calculations.
[0079] Where x1 represents the first dimension feature in the first Mel-Cepstral coefficient, x'1 represents the first dimension feature in the second Mel-Cepstral coefficient, x2 represents the second dimension feature in the first Mel-Cepstral coefficient, and x'2 represents the second dimension feature in the second Mel-Cepstral coefficient. n x' represents the nth dimension feature in the first Mel-Cepstral coefficients. n d represents the nth dimension feature in the second Mel-Cepstral coefficients, K represents a constant, and d represents the distance to the cluster center.
[0080] The clustering method based on Mel-Cepstral Coefficients described above has the advantage of fully considering the features of each dimension of the Mel-Cepstral Coefficients, thus obtaining the best clustering results.
[0081] Step S1620: Label the second Mel-Cepstral coefficients that are less than the first distance threshold from the cluster center as the type of the corresponding cluster center.
[0082] In one embodiment, a second Mel-Cepstral coefficient that is less than a first distance threshold of 0.3 from the cluster center is labeled as the type of the corresponding cluster center.
[0083] Step S1630: Perform rapid independent component analysis on the second Mel-Cepstral coefficients that are more than the second distance threshold from the cluster center, and label them as the corresponding types according to the analysis results.
[0084] In one embodiment, the second Mel-Cepstral coefficients that are more than 0.7 away from the cluster center are subjected to rapid independent component analysis, and are labeled as corresponding types according to the analysis results.
[0085] This embodiment uses clustering algorithms and fast independent component analysis algorithms to label the types of the second Mel-Cepstral coefficients, thus solving the problem that existing separation algorithms can only separate one signal at a time and are prone to aliasing when separating features in the second Mel-Cepstral coefficients.
[0086] Step S170: Train the initial detection model using the labeled second Mel-Cepstral coefficients to obtain the final detection model.
[0087] In this step, the labeled second Mel-frequency cepstral coefficients are input into the initial detection model for training. By calculating the error between the output value and the predicted value, the corresponding parameters of the initial detection model that do not meet the minimum error condition are adaptively adjusted until the minimum error condition is met, thus obtaining the final detection model.
[0088] In one embodiment, a portion of the second Mel-Cepstral coefficients can be used as a second training set, and another portion as a second test set. The second training set is used to train the initial detection model to obtain a first final detection model, and the second test set is used to test the first final detection model to obtain a second final detection model, i.e., the final detection model.
[0089] Step S180: Detect the acoustic fingerprint information of the transformer according to the final detection model to identify the transformer fault type.
[0090] Finally, the acoustic signature information of the transformer is detected using the final detection model to obtain efficient and accurate transformer fault type prediction results.
[0091] In summary, the transformer acoustic signature detection method of this application collects first acoustic signature information of the transformer under ideal operating conditions, and trains a preset neural network to obtain an initial detection model based on the first Mel-Cepstral coefficients obtained by feature extraction of the first acoustic signature information. Then, it collects second acoustic signature information of the transformer under complex environments, and performs corresponding type labeling on the second Mel-Cepstral coefficients extracted from the second acoustic signature information. The type-labeled second Mel-Cepstral coefficients are then input into the initial detection model to train a final detection model. The final detection model is used to detect the acoustic signature information to be tested, and finally, the fault type of the transformer is identified efficiently and accurately.
[0092] Based on the same inventive concept, this application also provides a transformer acoustic signature detection device.
[0093] Figure 4 Here is a block diagram of the transformer acoustic signature detection device, as shown below. Figure 4 As shown, the device may include the following modules:
[0094] The first acquisition module 210 is configured to acquire the first acoustic fingerprint information of the transformer, which is the sound information of the transformer under ideal operating conditions.
[0095] The first feature extraction module 220 is configured to extract features from the first voiceprint information to obtain the first Mel-frequency cepstral coefficients.
[0096] The first training module 230 is configured to input the first Mel-Cepstral coefficients into a preset neural network for training to obtain an initial detection model;
[0097] The second acquisition module 240 is configured to acquire the second acoustic fingerprint information of the transformer, which is the sound information of the transformer in a complex environment.
[0098] The second feature extraction module 250 is configured to extract features from the second voiceprint information to obtain the second Mel-frequency cepstral coefficients.
[0099] The type labeling module 260 is configured to label the type of the second Mel-frequency cepstral coefficients using clustering algorithms and fast independent component analysis algorithms;
[0100] The second training module 270 is configured to train the initial detection model using the labeled second Mel-Cepstral coefficients to obtain the final detection model.
[0101] The detection module 280 is configured to detect the acoustic fingerprint information of the transformer to be tested according to the final detection model, so as to realize the identification of the transformer fault type.
[0102] Please refer to Figure 5The diagram below is a device block diagram of a type labeling module in one embodiment of this application, wherein the type labeling module 260 further includes the following sub-modules:
[0103] The central setting submodule 2610 is configured to perform clustering calculations on the second Mel-Cepstral Coefficients using the first Mel-Cepstral Coefficients as the cluster center;
[0104] The first annotation submodule 2620 is configured to annotate the second Mel-Cepstral coefficients whose distance from the cluster center is less than a first distance threshold as the type of the corresponding cluster center;
[0105] The second annotation submodule 2630 is configured to perform rapid independent component analysis on the second Mel-Cepstral coefficients that are more than a second distance threshold from the cluster center, and to annotate them as the corresponding types according to the analysis results.
[0106] The specific implementation process of the functions and roles of each module in the above-mentioned device can be found in the implementation process of the corresponding steps in the above-mentioned transformer acoustic fingerprint detection method, and will not be repeated here.
[0107] Based on the same inventive concept, embodiments of this application also provide an electronic device.
[0108] Figure 6 This is a device block diagram of an electronic device provided in one embodiment of this application.
[0109] Please refer to Figure 6 The electronic device 300 includes: at least one processor 310 and a memory 320. Figure 5 Taking a processor 310 as an example, the processor 310 and the memory 320 are connected via a bus 330. The memory 320 stores instructions that can be executed by the processor 310. The instructions are executed by the processor 310 so that the electronic device 300 can execute all or part of the process of the transformer acoustic fingerprint detection method in the above embodiment.
[0110] The memory 320 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.
[0111] It should be noted that the electronic device 300 is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.
[0112] This application also provides a computer-readable storage medium storing a computer program that can be executed by a processor 310 to perform the transformer acoustic fingerprint detection method provided in this application.
[0113] The apparatuses and methods disclosed in the several embodiments provided in this application can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatuses, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
[0114] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.
[0115] It should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.
Claims
1. A transformer voiceprint detection method, characterized in that, include: Collect the first acoustic fingerprint information of the transformer, which is the sound information of the transformer under ideal operating conditions; Feature extraction is performed on the first voiceprint information to obtain the first Mel-frequency cepstral coefficients; The first Mel-frequency cepstral coefficients are input into a preset neural network for training to obtain an initial detection model; Collect the second acoustic fingerprint information of the transformer, which is the sound information of the transformer in a complex environment; Feature extraction is performed on the second voiceprint information to obtain the second Mel-frequency cepstral coefficients; Clustering algorithms and fast independent component analysis algorithms were used to label the types of the second Mel-frequency cepstral coefficients; The initial detection model is trained using the labeled second Mel-Cepstral coefficients to obtain the final detection model; The acoustic signature information of the transformer to be tested is detected based on the final detection model to identify the transformer fault type; The specific steps of labeling the type of the second Mel-Cepstral coefficient using clustering algorithms and fast independent component analysis algorithms include: using the first Mel-Cepstral coefficient as the cluster center, performing clustering calculations on the second Mel-Cepstral coefficient; labeling the second Mel-Cepstral coefficients whose distance from the cluster center is less than a first distance threshold as the type of the corresponding cluster center; and performing fast independent component analysis on the second Mel-Cepstral coefficients whose distance from the cluster center is greater than a second distance threshold, and labeling them as the corresponding types according to the analysis results. Before labeling the type of the second Mel-Cepstral Coefficient using clustering and fast independent component analysis algorithms, the method further includes: scoring the second Mel-Cepstral Coefficient by calculating the error between its true value and its predicted value. The second Mel-Cepstral Coefficient with a score lower than the first score is used for type labeling in subsequent clustering and fast independent component analysis algorithms, while the second Mel-Cepstral Coefficient with a score higher than the second score is used to train the initial detection model.
2. The transformer voiceprint detection method of claim 1, wherein, The step of extracting features from the first voiceprint information to obtain the first Mel-spectral coefficients specifically includes: The first voiceprint information is subjected to high-pass filtering to remove low-frequency components from the first voiceprint information. The first voiceprint information, after high-pass filtering, is then processed into frames. Windowing is applied to the first audioprint information after framing. The first voiceprint information after windowing is subjected to time-frequency transformation; Extract the frequency response of the first voiceprint information after time-frequency transformation; The first Mel-frequency cepstral coefficients are obtained by performing logarithmic and cosine transformations on the frequency response.
3. The transformer acoustic signature detection method according to claim 1, characterized in that, The clustering calculation of the second Mel-Cepstral Coefficients using the first Mel-Cepstral Coefficient as the cluster center specifically includes, through the formula: ； ； Clustering calculations are performed, where, x 1 represents the first dimension feature in the first Mel-Cepstral coefficients. x’ 1 represents the first dimension feature in the second Mel-Cepstral coefficients. x 2 represents the second dimension feature in the first Mel-frequency cepstral coefficients. x’ 2 represents the second dimension feature in the second Mel-frequency cepstral coefficients. x n This represents the nth dimension feature in the first Mel-frequency cepstral coefficients. x’ n d represents the nth dimension feature in the second Mel-Cepstral coefficients, K represents a constant, and d represents the distance to the cluster center.
4. The transformer acoustic signature detection method according to claim 3, characterized in that, The first distance threshold is 0.3, and the second distance threshold is 0.
7.
5. A transformer voiceprint detection device, characterized in that, The apparatus is used to perform the method according to any one of claims 1 to 4, and the apparatus comprises the following modules: The first acquisition module is configured to acquire the first acoustic fingerprint information of the transformer, which is the sound information of the transformer under ideal operating conditions. The first feature extraction module is configured to extract features from the first voiceprint information to obtain the first Mel-frequency cepstral coefficients. The first training module is configured to input the first Mel-Cepstral coefficients into a preset neural network for training to obtain an initial detection model; The second acquisition module is configured to acquire the second acoustic fingerprint information of the transformer, which is the sound information of the transformer in a complex environment. The second feature extraction module is configured to extract features from the second voiceprint information to obtain the second Mel-frequency cepstral coefficients. The type labeling module is configured to label the type of the second Mel-frequency cepstral coefficients using clustering algorithms and fast independent component analysis algorithms; The second training module is configured to train the initial detection model using the labeled second Mel-Cepstral coefficients to obtain the final detection model. The detection module is configured to detect the acoustic fingerprint information of the transformer to be tested according to the final detection model, so as to identify the transformer fault type.
6. The transformer acoustic signature detection device according to claim 5, characterized in that, The type annotation module includes: The central setting submodule is configured to perform clustering calculations on the second Mel-Cepstral Coefficients using the first Mel-Cepstral Coefficient as the cluster center; The first annotation submodule is configured to annotate the second Mel-Cepstral coefficients that are less than a first distance threshold from the cluster center as the type of the corresponding cluster center; The second annotation submodule is configured to perform rapid independent component analysis on the second Mel-Cepstral coefficients that are more than a second distance threshold from the cluster center, and to annotate them as the corresponding types according to the analysis results.
7. An electronic device, characterized in that, The electronic device includes: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute the transformer acoustic signature detection method according to any one of claims 1 to 4.
8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, which can be executed by one or more processors to implement the transformer acoustic print detection method as described in any one of claims 1 to 4.