Transformer state detection method and device

Abstract

The invention provides a transformer state detection method and device. The method comprises the following steps: acquiring voiceprint information when a transformer runs; inputting the voiceprint information into a recognition model to obtain a recognition result output by the recognition model, the recognition model being a neural network model obtained according to sample voiceprint training, and the sample voiceprint being a voiceprint with a fault information tag extracted from the voiceprint information; and outputting fault judgment information according to the identification result. Whether the transformer is normal or not is judged by collecting the voiceprint information of the operation of the transformer, and after the sound of a certain fault is calibrated, if the sound with the same characteristic is encountered again, the corresponding fault can be directly recognized, so that the problem that the fault of the transformer is difficult to detect is solved.

Description

Technical Field

[0001] This application relates to the technical field of substation operation, and specifically relates to a transformer status detection method and device. Background Art

[0002] As a static electrical appliance, a transformer is mainly used for power grid systems, long-distance power transmission, etc. During the operation of the transformer, if there are internal faults, there will be light emission, heat generation, and strange smells. However, due to the influence of the transformer shell, these physical phenomena are not easily directly detected. However, some physical phenomena can be transmitted through the transformer shell, such as sound waves, vibrations, etc.

[0003] There are many methods for detecting transformer abnormalities. For example, the intuitive detection method can quickly detect transformer faults, but it requires experienced personnel for detection, otherwise there may be missed detections or misdiagnoses. The electrical preventive test method can detect local discharges and insulation aging problems inside the transformer. The dissolved gas analysis in oil method can detect overheating problems inside the transformer, but it requires professional equipment and technical personnel for operation and analysis, resulting in difficulties in transformer fault detection. Summary of the Invention

[0004] This application provides a transformer status detection method and device to solve the problem of difficult transformer fault detection.

[0005] In a first aspect, this application provides a transformer status detection method, including:

[0006] Obtain the voiceprint information of the transformer during operation;

[0007] Input the voiceprint information into an identification model to obtain the identification result output by the identification model. The identification model is a neural network model trained according to sample voiceprints, and the sample voiceprints are voiceprints extracted from the voiceprint information with fault information labels; according to the identification result, output fault determination information.

[0008] Optionally, obtaining the voiceprint information of the transformer during operation includes:

[0009] Collect the sound information of the transformer during operation through a microphone. The sound information is an electrical signal generated based on sound;

[0010] Perform voiceprint feature recognition and extraction on the sound information to obtain voiceprint information; the feature is a short-time frequency spectrum feature.

[0011] Optionally, collecting the sound information of the transformer during operation through a microphone includes:

[0012] Obtain the noise information of the environment where the transformer is located;

[0013] Perform a short-time Fourier transform on the noise information to obtain a noise feature voiceprint;

[0014] Compare the voice information with the noise feature voiceprint;

[0015] If the coincidence degree between the voice information and the noise feature voiceprint is greater than a preset coincidence degree threshold, obtain voice information again.

[0016] Optionally, the method further includes:

[0017] Construct a data set according to the voiceprint information;

[0018] Construct an identification model based on a neural network, where the neural network includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer connected in sequence;

[0019] Input the data set into the identification model for model training.

[0020] Optionally, before the step of constructing a data set according to the voiceprint information, it further includes:

[0021] Add fault information labels to a number of the voiceprint information to construct a data set with label information;

[0022] Divide the data set into a training set, a validation set, and a test set; the training set is used to construct an identification model, the validation set is used to verify the identification model, and the test set is used to test the identification model.

[0023] Optionally, the step of inputting the data set into the identification model for model training includes:

[0024] Pre-train the identification model using the voiceprint information in the training set and calculate the generated loss;

[0025] If the generated loss is less than or equal to a preset loss threshold,

[0026] Then use the identification model to transfer to the validation set for retraining, and adjust the parameters of the identification model according to the identification result;

[0027] Use the identification model to transfer to the test set for verification and test the accuracy of the identification result;

[0028] If the accuracy is less than a preset accuracy threshold, save the identification model;

[0029] If the generated loss is greater than a preset loss threshold, adjust the model parameters of the identification module according to the generated loss, and continue iterative training based on the identification model after adjusting the model parameters.

[0030] Optionally, characterized in that, according to the recognition result, the output fault determination information includes:

[0031] If the recognition result is a single fault type, output the fault determination information including the fault type of the transformer;

[0032] If the recognition result is multiple fault types, output the fault determination information including the fault type probability of the transformer.

[0033] In a second aspect, the present application provides a transformer status detection device, which is applied to the transformer status detection method provided in the first aspect above, and includes: a device status detection component and a monitoring host;

[0034] The device status detection component is communicatively connected to the monitoring host;

[0035] The device status detection component is configured to: obtain the voiceprint information during the operation of the transformer;

[0036] The monitoring host is configured to:

[0037] Obtain the voiceprint information during the operation of the transformer;

[0038] Input the voiceprint information into the recognition model to obtain the recognition result output by the recognition model. The recognition model is a neural network model trained according to the sample voiceprint, and the sample voiceprint is the voiceprint with a fault information label extracted from the voiceprint information;

[0039] According to the recognition result, output the fault determination information.

[0040] Optionally, it further includes a power supply module. The monitoring host further includes a first aviation socket, a signal receiving antenna and a display unit. The device status detection component includes a second aviation socket. The power supply module is respectively connected to the first aviation socket and the second aviation socket;

[0041] The signal receiving antenna is connected to the monitoring host, and the display unit is connected to the monitoring host;

[0042] The signal receiving antenna is configured to: receive the electrical signal sent by the device status detection component and send the electrical signal to the monitoring host;

[0043] The monitoring host is configured to: obtain the electrical signal and perform feature recognition and extraction on the electrical signal to obtain the voiceprint information;

[0044] The display unit is configured to: display the voiceprint information and the fault determination.

[0045] Optionally, the device status detection component includes a magnetic adsorption part, a microphone, and a synchronous acquisition card;

[0046] The magnetic adsorption part is used to: adsorb the device status detection component on the transformer;

[0047] The microphone is used to: collect the sound information during the operation of the transformer, and the sound information is an electrical signal generated based on the sound;

[0048] The synchronous acquisition card is used to: send the electrical signal to the signal receiving antenna.

[0049] As can be seen from the above technical solutions, the present application provides a transformer status detection method and device. The method includes: obtaining the voiceprint information during the operation of the transformer; inputting the voiceprint information into an identification model to obtain the identification result output by the identification model. The identification model is a neural network model trained according to sample voiceprints, and the sample voiceprints are voiceprints with fault information labels extracted from the voiceprint information; according to the identification result, output fault determination information. By collecting the voiceprint information during the operation of the transformer to determine whether the transformer is normal, when the sound of a certain fault is calibrated, if the sound with the same characteristics is encountered again, the corresponding fault can be directly identified to solve the problem of difficult transformer fault detection. Description of the Drawings

[0050] In order to more clearly illustrate the technical solutions of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0051] Figure 1 It is a flowchart of the transformer status detection method provided by the embodiment of the present application;

[0052] Figure 2 It is a flowchart of obtaining the voiceprint information of the transformer provided by the embodiment of the present application;

[0053] Figure 3 It is a flowchart of constructing the identification model provided by the embodiment of the present application;

[0054] Figure 4 It is a schematic diagram of the structure of the neural network model provided by the embodiment of the present application;

[0055] Figure 5 It is a schematic diagram of the structure of the monitoring host provided by the embodiment of the present application;

[0056] Figure 6 It is a schematic diagram of the structure of the device status detection component provided by the embodiment of the present application.

[0057] Reference Signs:

[0058] Wherein: 101 - first aviation socket; 102 - display unit; 103 - signal receiving antenna; 201 - second aviation socket; 202 - magnetic attraction part. Specific embodiments

[0059] Embodiments will be described in detail below, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following examples do not represent all embodiments consistent with the present application. They are merely examples of systems and methods consistent with some aspects of the present application detailed in the claims.

[0060] To solve the problem of difficult detection of transformer faults, refer to Figure 1 , some embodiments of the present application provide a transformer status detection method, including:

[0061] S10: Obtain the voiceprint information of the transformer during operation.

[0062] Wherein, as Figure 2 shown, obtaining the voiceprint information of the transformer during operation specifically includes:

[0063] S100: Collect the sound information of the transformer during operation through a microphone, and the sound information is an electrical signal generated based on the sound.

[0064] To obtain the voiceprint information of the transformer during operation, first, the sound information of the transformer during operation can be collected through a microphone. Select a suitable microphone and collection device, and determine the placement position and quantity of the collection device. At the same time, it is necessary to select a suitable preprocessing method and technology according to the actual situation to remove noise and other interference factors.

[0065] For example, in the environment where the transformer operates, the sound of the transformer body operation and the noise in the environment will be collected at the same time. To ensure the accuracy of the detection, it is necessary to determine whether the collected sound information can be used. Among them, determining whether the collected sound information can be used includes: obtaining the noise information of the environment where the transformer is located; performing a short-time Fourier transform on the noise information to obtain the noise characteristic voiceprint; comparing the sound information with the noise characteristic voiceprint; if the coincidence degree between the sound information and the noise characteristic voiceprint is greater than the preset coincidence degree threshold, re-obtain the sound information.

[0066] First, it is necessary to first collect the ambient noise around the transformer, and then use the short-time Fourier transform algorithm to convert the ambient noise into a two-dimensional time-frequency spectrum from a one-dimensional signal through frame division, windowing, and discrete Fourier transform. The windowing calculation formula is as follows:

[0067]

[0068] Performing a short-time discrete Fourier transform on the time-domain frames after discretization yields the time-frequency spectrum matrix. The formula for the time-frequency spectrum matrix is as follows:

[0069] 0 ≤ n and k ≤ N - 1;

[0070] where K is the frequency point index, and X(m, n) and X(m, k) are the original discrete time-domain signal of the m-th frame and the frequency spectrum distribution of the m-th frame, respectively.

[0071] After obtaining the original discrete time-domain signal and the frequency spectrum distribution of the ambient noise around the transformer, the noise characteristic voiceprint can be obtained. Then, the sound information acquired by the microphone is compared with the noise characteristic voiceprint. If the coincidence degree between the sound information and the noise characteristic voiceprint is greater than the preset coincidence degree threshold, it indicates that the recorded sound information belongs to the ambient sound around the device, and the sound needs to be recorded again. If the coincidence degree between the sound information and the noise characteristic voiceprint is less than the preset coincidence degree threshold, the sound information can be used as the basis for determining whether the transformer is faulty. By determining whether the recorded sound information belongs to the faulty sound of the transformer or the ambient sound around the transformer, the detection accuracy can be improved.

[0072] S200: Perform voiceprint feature recognition and extraction on the sound information to obtain voiceprint information.

[0073] Voiceprint information is a type of characteristic information with uniqueness and non-replicability. Therefore, by performing voiceprint feature recognition and extraction on the sound signal, voiceprint information can be obtained. Furthermore, whether the transformer is faulty can be determined through the voiceprint information, and the type of fault of the transformer can be determined through the voiceprint information.

[0074] In this embodiment, the voiceprint feature is the short-time frequency spectrum feature. The short-time frequency spectrum feature is an important feature of voiceprint information, which reflects the energy distribution of the sound signal at different frequencies. By extracting the short-time frequency spectrum feature of the sound signal, a set of feature vectors can be obtained, and this set of feature vectors can be used to represent the unique features of the sound signal, thereby performing transformer fault determination.

[0075] During the process of voiceprint feature extraction, it is necessary to process the sound signal, such as noise removal, signal enhancement, etc., to improve the accuracy and stability of feature extraction. In addition, appropriate feature extraction algorithms, such as MFCC (Mel Frequency Cepstrum Coefficient) and LPCC (Linear Prediction Cepstrum Coefficient), can be used to obtain more accurate and stable feature vectors.

[0076] S20: Input the voiceprint information into the recognition model to obtain the recognition result output by the recognition model.

[0077] Among them, before inputting the voiceprint information into the recognition model, it is necessary to construct a recognition model based on a neural network. Among them, as Figure 3 shown, the specific steps for constructing the recognition model include:

[0078] S201: Construct a data set according to the voiceprint information.

[0079] Among them, the specific steps for constructing a data set according to the voiceprint information include: adding fault information labels to a number of voiceprint information to construct a data set with label information. Divide the data set into a training set, a validation set, and a test set.

[0080] Specifically, to add fault information labels to the voiceprint information, it is first necessary to collect voiceprint data with fault information. This data can be obtained through various channels, such as maintenance records, fault reports, on-site collection, etc. When collecting data, it is necessary to record the fault types related to the voiceprint information, and add labels to the voiceprint data according to the collected fault information. The labels can be fault types, fault levels, etc., depending on the application scenario and requirements. For each voiceprint data, it is necessary to associate it with the corresponding fault label.

[0081] By constructing a data set with voiceprint information with fault information labels, and then dividing the data set into a training set, a validation set, and a test set. Among them, the training set is the data set used to construct the recognition model, usually accounting for 60%-80% of the entire data set. The role of the training set is to train the parameters of the model so that the model can learn the characteristics and laws of the data, so as to be able to predict and classify new data.

[0082] The validation set is used to verify the recognition model, usually accounting for 10%-20% of the entire data set. The role of the validation set is to evaluate and adjust the parameters of the model during the model training process. Through the validation set, we can understand the performance of the model on unknown data and optimize and improve the model.

[0083] The test set is used to test the recognition model, usually accounting for 10%-20% of the entire data set. The role of the test set is to evaluate the final model after the model training and optimization. Through the test set, we can understand the performance of the model in the real scenario and evaluate the accuracy and reliability of the model.

[0084] During the process of dividing the dataset, the ratios of the training set, validation set, and test set should be adjusted according to the actual situation. If the dataset is very large, the ratio of the training set can be appropriately reduced; otherwise, it should be increased. At the same time, attention should be paid to randomness during the dataset division process: the dataset should be divided by random sampling to avoid the influence of human factors on the results. When the dataset is small, cross-validation can be used for model training and validation to increase the utilization rate of the data. In short, the division of the dataset is a very important step in machine learning and deep learning. By reasonably dividing the dataset, an accurate and reliable recognition model can be obtained.

[0085] S202: Build a recognition model based on a neural network.

[0086] The convolutional neural network in the neural network algorithm is a deep learning algorithm that supports multi-layer neural networks and can simultaneously learn and optimize feature parameters and classifiers. As Figure 4 shown, the neural network includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer connected in sequence. After building a recognition model based on a neural network, it is necessary to establish a voiceprint information database by collecting data on different voiceprint features of transformers, and then train the recognition model so that the recognition model can autonomously learn various voiceprint features and make corresponding matches with various transformer faults using the neural network algorithm to improve the fault detection accuracy.

[0087] S203: Input the dataset into the recognition model for model training.

[0088] Among them, during the model training process, first pre-train the recognition model using the voiceprint information in the training set and calculate the generated loss; if the generated loss is less than or equal to the preset loss threshold, then use the recognition model to transfer to the validation set for re-training and adjust the parameters of the recognition model according to the recognition results. Use the recognition model to transfer to the test set for verification and test the accuracy of the recognition results; if the accuracy is less than the preset accuracy threshold, then save the recognition model; if the generated loss is greater than the preset loss threshold, then adjust the model parameters of the recognition module according to the generated loss, and continue iterative training based on the recognition model after adjusting the model parameters.

[0089] For example, obtain 1000 voiceprint information, and then divide the 1000 voiceprint information into a training set, a test set, and a validation set. Use the training set for pre-training: at the beginning stage of model training, pre-train the recognition model using the voiceprint information in the training set. The goal of this stage is to enable the recognition model to learn the ability to extract features from voiceprints and recognize them.

[0090] After each training iteration, the recognition model generates prediction results and evaluates the performance of the recognition model by calculating the loss between the prediction results and the true labels. Loss is a metric for measuring the degree of prediction error. The greater the difference between the prediction results and the true labels, the greater the loss. If the generated loss is less than or equal to the preset loss threshold, it means that the prediction performance of the model has reached an acceptable level and the next step can be entered.

[0091] If the generated loss of the recognition model is lower than the preset loss threshold, then the recognition model is transferred to the validation set for retraining. This step is to further optimize the performance of the recognition model on a new dataset. On the validation set, the parameters of the model are fine-tuned according to the recognition results of the recognition model. This step is to make the recognition model better adapt to the new data distribution.

[0092] The adjusted model is used to validate on the test set to test the accuracy of the recognition model. If the accuracy is lower than the preset accuracy threshold, it means that the performance of the recognition model still needs to be improved.

[0093] If the accuracy of the recognition model reaches the preset accuracy threshold, then the current recognition model can be saved as the final recognition model. Multiple recognition models are trained to obtain a recognition model that meets the requirements.

[0094] S30: Output fault determination information according to the recognition result.

[0095] If the recognition result is a single fault type, output fault determination information including the fault type of the transformer;

[0096] If the recognition result is multiple fault types, output fault determination information including the fault type probabilities of the transformer.

[0097] If there is only one identified fault type of the transformer in the recognition result, then fault determination information including that fault type will be output. For example, if there is a fault in the winding, the system will output fault determination information of "winding fault". If there is more than one fault type, then fault determination information including each fault type and its probability will be output. For example, if there are faults in both the winding and the iron core, it will output fault determination information of "winding fault (0.6), iron core fault (0.4)".

[0098] While outputting the fault determination information, the system will also give corresponding handling suggestions according to the type and degree of the fault. For example, if it is a winding fault, the system will suggest repairing or replacing the winding; if it is a core fault, the system will suggest repairing or replacing the core. By collecting the acoustic fingerprint information of the transformer during operation to determine whether the transformer is normal, when the sound of a certain fault is calibrated, if the sound with the same characteristics is encountered again, the corresponding fault can be directly identified to solve the problem of difficult fault detection of the transformer.

[0099] In some embodiments, such as Figure 5 , Figure 6 shown, some embodiments of the present application further provide a transformer status detection device, including: a device status detection component and a monitoring host;

[0100] The device status detection component is communicatively connected to the monitoring host;

[0101] The device status detection component is configured to: obtain the acoustic fingerprint information of the transformer during operation;

[0102] The monitoring host is configured to:

[0103] Obtain the acoustic fingerprint information of the transformer during operation;

[0104] Input the acoustic fingerprint information into the recognition model to obtain the recognition result output by the recognition model. The recognition model is a neural network model trained according to the sample acoustic fingerprints, and the sample acoustic fingerprints are the acoustic fingerprints with fault information labels extracted from the acoustic fingerprint information;

[0105] According to the recognition result, output the fault determination information.

[0106] The acoustic fingerprint information is the unique sound feature of the transformer during operation, which can reflect the operating state of the transformer. The device status detection component collects the acoustic fingerprint information of the transformer through devices such as sensors and microphones and transmits it to the monitoring host. The monitoring host inputs the received acoustic fingerprint information into a neural network model trained with sample acoustic fingerprints. This neural network model is trained by learning a large amount of acoustic fingerprint data with fault information labels, and it can automatically identify the acoustic fingerprint features in normal and fault situations. Then, the monitoring host determines whether the transformer has a fault according to the output result of the neural network model and outputs the corresponding fault determination information.

[0107] The transformer status detection device based on acoustic fingerprint information can monitor the operating state of the transformer in real time, discover and warn of faults in time, effectively avoid the sudden damage of the transformer and accidents, and improve the stability and reliability of the power system. At the same time, this method can also achieve remote monitoring and unattended operation, which is convenient for power workers to manage and maintain anytime and anywhere.

[0108] In some embodiments, the transformer status detection device further includes a power supply module, the monitoring host further includes a first aviation socket 101, a signal receiving antenna 102, and a display unit 103, and the device status detection component includes a second aviation socket 201. The power supply module is respectively connected to the first aviation socket 101 and the second aviation socket 201;

[0109] The signal receiving antenna 102 is connected to the monitoring host, and the display unit 103 is connected to the monitoring host;

[0110] The signal receiving antenna 102 is configured to: receive the electrical signal sent by the device status detection component and send the electrical signal to the monitoring host;

[0111] The monitoring host is configured to: obtain the electrical signal and perform feature recognition and extraction on the electrical signal to obtain voiceprint information;

[0112] The display unit 103 is configured to: display the voiceprint information and the fault determination.

[0113] In some embodiments, the device status detection component includes a magnetic attraction part 202, a microphone 203, and a synchronous acquisition card;

[0114] The magnetic attraction part 202 is used to: adsorb the device status detection component on the transformer;

[0115] The microphone is used to: collect the sound information during the operation of the transformer, and the sound information is an electrical signal generated based on the sound;

[0116] The synchronous acquisition card is used to: send the electrical signal to the signal receiving antenna 102.

[0117] The magnetic attraction part can firmly adsorb the device status detection component on the transformer, thereby ensuring that the device status detection component can stably collect the sound information during the operation of the transformer.

[0118] The microphone is used to collect the sound information during the operation of the transformer. These sound information exist in the form of electrical signals. That is to say, what the microphone collects is the electrical signal generated based on the sound. These electrical signals contain the operation status information of the transformer, such as whether there is abnormal noise and whether there is a fault. The synchronous acquisition card is then used to send the electrical signal to the signal receiving antenna 102. During the sending process, the synchronous acquisition card can also encode and encrypt the electrical signal to ensure the stability and security of the signal. Through the collaborative work of these components, the device status detection component can monitor the operation status of the transformer in real time, detect abnormal situations in a timely manner, and provide a strong guarantee for the stable operation of the power system.

[0119] As can be seen from the above technical solutions, the present application provides a transformer status detection method and device. The method includes: obtaining the voiceprint information of the transformer during operation; inputting the voiceprint information into an identification model to obtain the identification result output by the identification model, where the identification model is a neural network model trained according to sample voiceprints, and the sample voiceprints are voiceprints with fault information labels extracted from the voiceprint information; and outputting fault determination information according to the identification result. By collecting the voiceprint information of the transformer during operation to determine whether the transformer is normal, when the sound of a certain fault is calibrated, if the sound with the same characteristics is encountered again, the corresponding fault can be directly identified to solve the problem of difficult transformer fault detection.

[0120] For the similar parts between the embodiments provided in the present application, reference can be made to each other. The specific embodiments provided above are only several examples under the general concept of the present application and do not constitute a limitation on the protection scope of the present application. For those skilled in the art, any other implementation manner extended based on the solution of the present application without creative efforts belongs to the protection scope of the present application.

Claims

1. A transformer status detection method, characterized in that, Including: Obtain the voiceprint information of the transformer during operation; Input the voiceprint information into an identification model to obtain the identification result output by the identification model. The identification model is a neural network model trained according to sample voiceprints, and the sample voiceprints are voiceprints with fault information labels extracted from the voiceprint information; Output fault determination information according to the identification result.

2. The transformer status detection method according to claim 1, characterized in that, Obtaining the voiceprint information of the transformer during operation includes: Collect the sound information of the transformer during operation through a microphone. The sound information is an electrical signal generated based on sound; Perform voiceprint feature recognition and extraction on the sound information to obtain voiceprint information; the feature is a short-time spectrum feature.

3. The transformer status detection method according to claim 2, characterized in that, Collecting the sound information of the transformer during operation through a microphone includes: Obtain the noise information of the environment where the transformer is located; Perform short-time Fourier transform on the noise information to obtain a noise feature voiceprint; Compare the sound information with the noise feature voiceprint; If the coincidence degree between the sound information and the noise feature voiceprint is greater than a preset coincidence degree threshold, re-obtain the sound information.

4. The transformer status detection method according to claim 1, characterized in that, The method further includes: Construct a data set according to the voiceprint information; Construct an identification model based on a neural network. The neural network includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer connected in sequence; Input the data set into the identification model for model training.

5. The transformer status detection method according to claim 4, wherein Before the step of constructing the data set according to the voiceprint information, it further includes: Add fault information labels to several pieces of the voiceprint information to construct a data set with label information; Divide the data set into a training set, a validation set, and a test set; the training set is used to construct the identification model, the validation set is used to verify the identification model, and the test set is used to test the identification model.

6. The transformer status detection method according to claim 5, wherein The step of inputting the data set into the identification model for model training includes: Pre-train the identification model using the voiceprint information in the training set and calculate the generated loss; If the generated loss is less than or equal to a preset loss threshold, Then use the identification model to migrate to the validation set for retraining, and adjust the parameters of the identification model according to the identification result; Use the identification model to migrate to the test set for verification and test the accuracy of the identification result; If the accuracy is less than a preset accuracy threshold, save the identification model; If the generated loss is greater than a preset loss threshold, adjust the model parameters of the identification module according to the generated loss, and continue iterative training based on the identification model with adjusted model parameters.

7. The transformer status detection method according to claim 1, characterized in that Outputting fault determination information according to the identification result includes: If the identification result is a single fault type, output the fault determination information including the fault type of the transformer; If the identification result is multiple fault types, output the fault determination information including the fault type probability of the transformer.

8. A transformer status detection device, characterized in that Applied to the transformer status detection method according to any one of claims 1-7, including: a device status detection component and a monitoring host; The device status detection component is communicatively connected to the monitoring host; The device status detection component is configured to: obtain the voiceprint information during the operation of the transformer; The monitoring host is configured to: Obtain the voiceprint information during the operation of the transformer; Input the voiceprint information into an identification model to obtain the identification result output by the identification model. The identification model is a neural network model trained based on sample voiceprints, and the sample voiceprints are voiceprints with fault information labels extracted from the voiceprint information; Output fault determination information according to the identification result.

9. The transformer status detection device according to claim 8, characterized in that It further includes a power supply module. The monitoring host further includes a first aviation socket, a signal receiving antenna, and a display unit. The device status detection component includes a second aviation socket. The power supply module is respectively connected to the first aviation socket and the second aviation socket; The signal receiving antenna is connected to the monitoring host, and the display unit is connected to the monitoring host; The signal receiving antenna is configured to: receive the electrical signal sent by the device status detection component and send the electrical signal to the monitoring host; The monitoring host is configured to: obtain the electrical signal and perform feature recognition and extraction on the electrical signal to obtain voiceprint information; The display unit is configured to: display the voiceprint information and the fault determination.

10. The transformer status detection device according to claim 8, characterized in that, The device status detection component includes a magnetic adsorption part, a microphone, and a synchronous acquisition card; The magnetic adsorption part is used to: adsorb the device status detection component on the transformer; The microphone is used to: collect the sound information during the operation of the transformer, and the sound information is an electrical signal generated based on the sound; The synchronous acquisition card is used to: send the electrical signal to the signal receiving antenna.

Images