Transformer abnormal voiceprint diagnosis system and method based on neural network

Abstract

The invention relates to the field of voiceprint diagnosis, and particularly discloses a transformer abnormal voiceprint diagnosis system and method based on a neural network. Firstly, a transformer winding thermogram collected by a thermal infrared imager, transformer winding vibration signals collected by a vibration sensor at a plurality of preset time points and transformer winding voiceprint signals collected by a voiceprint sensor at a plurality of preset time points are obtained, and then feature extraction and correlation analysis are carried out on the three parts by using a deep learning technology to obtain a transformer winding thermogram, a transformer winding vibration signal and a transformer winding voiceprint signal. And finally, a classification result is obtained through a classifier to judge whether the transformer winding has abnormal voiceprints or not, so that whether the transformer winding is deformed or not is judged, the problems of the transformer winding can be found as early as possible, corresponding repair and maintenance measures are taken, further damage and accidents are avoided, and the safety of the transformer winding is improved. And preventive maintenance and improvement of the reliability of the power grid are further achieved.

Description

Technical Field

[0001] The present application relates to the field of voiceprint diagnosis, and more specifically, to a transformer abnormal voiceprint diagnosis system and method based on neural network. Background Art

[0002] A transformer is an electrical device used to transfer AC power from one circuit to another, achieving voltage conversion through the principle of electromagnetic induction. It consists of two or more windings, including a main winding and a secondary winding. The main winding is responsible for transferring power from the input circuit to the output circuit, while the secondary winding is used to achieve different voltage conversion ratios or provide auxiliary functions. The winding is made of wire or coil, and the appropriate wire cross-sectional area and number of turns are selected according to the design requirements. The working principle of the transformer is to cause voltage changes in the winding through a changing magnetic field. The performance and efficiency of the transformer depend on the design of the winding, the choice of insulation material, and the manufacturing quality of the winding. Reasonable design, manufacture and maintenance of transformer windings are essential to ensure the performance and reliability of the transformer.

[0003] In recent years, with the increase in power system capacity and load, the electromagnetic environment of substations has become more and more complex, which has led to an increase in the probability of transformer failure. According to statistical data, an important cause of transformer failure is the deformation of the winding, and the main reason for the deformation of the winding is the unbalanced electromagnetic force generated in the winding after the external short circuit of the transformer. At present, the frequency response analysis method is a commonly used method for transformer winding deformation detection. It has the advantages of being easy to use and able to well reflect the internal mechanical condition of the transformer, so it is widely used. However, the use of frequency response curves to analyze the type and degree of transformer winding faults is heavily dependent on expert knowledge, and the diagnostic conclusions are highly subjective and may even vary from person to person, resulting in different diagnostic conclusions for the same frequency curve.

[0004] Therefore, a transformer abnormal soundprint diagnosis system and method based on neural network is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a transformer abnormal soundprint diagnosis system and method based on neural network, which first obtains the transformer winding thermal image acquired by the infrared thermal imager, the transformer winding vibration signal acquired by the vibration sensor at multiple predetermined time points, and the transformer winding soundprint signal acquired by the soundprint sensor at multiple predetermined time points, and then uses the deep learning technology to perform feature extraction and correlation analysis on the three, and finally obtains the classification result through the classifier to determine whether the transformer winding has abnormal soundprint, thereby determining whether the transformer winding has deformation, which helps to detect the transformer winding problem early, and take corresponding repair and maintenance measures to avoid further damage and accidents, and further achieve preventive maintenance and improve power grid reliability.

[0006] According to one aspect of the present application, a transformer abnormal voiceprint diagnosis system based on a neural network is provided, which includes:

[0007] A transformer winding data acquisition module, used to acquire a transformer winding thermal image acquired by an infrared thermal imager, transformer winding vibration signals acquired by a vibration sensor at multiple predetermined time points, and transformer winding voiceprint signals acquired by a voiceprint sensor at multiple predetermined time points;

[0008] A transformer winding data extraction module, used to extract a transformer winding thermal image enhancement feature vector and a transformer winding global feature vector from the transformer winding thermal image acquired by the infrared thermal imager, the transformer winding vibration signals at multiple predetermined time points acquired by the vibration sensor, and the transformer winding voiceprint signals at multiple predetermined time points acquired by the voiceprint sensor;

[0009] The transformer winding voiceprint abnormality judgment module is used to judge whether the transformer winding has abnormal voiceprint based on the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector.

[0010] According to another aspect of the present application, a transformer abnormal voiceprint diagnosis method based on a neural network is provided, which includes:

[0011] Acquire a thermal image of a transformer winding acquired by an infrared thermal imager, a vibration signal of the transformer winding acquired by a vibration sensor at a plurality of predetermined time points, and a voiceprint signal of the transformer winding acquired by a voiceprint sensor at a plurality of predetermined time points;

[0012] Extracting a transformer winding thermal image enhancement feature vector and a transformer winding global feature vector from the transformer winding thermal image acquired by the infrared thermal imager, the transformer winding vibration signals at multiple predetermined time points acquired by the vibration sensor, and the transformer winding voiceprint signals at multiple predetermined time points acquired by the voiceprint sensor;

[0013] Based on the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector, it is determined whether the transformer winding has an abnormal soundprint.

[0014] Compared with the prior art, the present application provides a transformer abnormal soundprint diagnosis system and method based on a neural network, which first obtains a thermal image of the transformer winding collected by an infrared thermal imager, a transformer winding vibration signal at multiple predetermined time points collected by a vibration sensor, and a transformer winding soundprint signal at multiple predetermined time points collected by a soundprint sensor, and then uses deep learning technology to perform feature extraction and correlation analysis on the three, and finally obtains a classification result through a classifier to determine whether the transformer winding has an abnormal soundprint, thereby determining whether the transformer winding has deformation, which helps to detect problems with the transformer winding early and take corresponding repair and maintenance measures to avoid further damage and accidents, thereby further achieving preventive maintenance and improving power grid reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 It is a block diagram of a transformer abnormal voiceprint diagnosis system based on a neural network according to an embodiment of the present application.

[0017] Figure 2 It is a block diagram of a transformer winding data extraction module in a transformer abnormal voiceprint diagnosis system based on a neural network according to an embodiment of the present application.

[0018] Figure 3 It is a block diagram of a transformer winding thermal image feature extraction unit in a transformer abnormal soundprint diagnosis system based on a neural network according to an embodiment of the present application.

[0019] Figure 4 It is a block diagram of a transformer winding voiceprint abnormality judgment module in a transformer abnormal voiceprint diagnosis system based on a neural network according to an embodiment of the present application.

[0020] Figure 5 The flowchart is a method for diagnosing abnormal transformer soundprint based on a neural network according to an embodiment of the present application.

[0021] Figure 6 is a block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0023] Exemplary Systems

[0024] Figure 1 FIG. 1 is a block diagram of a transformer abnormal voiceprint diagnosis system based on a neural network according to an embodiment of the present application. Figure 1 As shown, according to the embodiment of the present application, the transformer abnormal voiceprint diagnosis system 100 based on neural network includes: a transformer winding data acquisition module 110, which is used to obtain the transformer winding thermal image collected by the infrared thermal imager, the transformer winding vibration signal at multiple predetermined time points collected by the vibration sensor, and the transformer winding voiceprint signal at multiple predetermined time points collected by the voiceprint sensor; a transformer winding data extraction module 120, which is used to extract the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector from the transformer winding thermal image collected by the infrared thermal imager, the transformer winding vibration signal at multiple predetermined time points collected by the vibration sensor, and the transformer winding voiceprint signal at multiple predetermined time points collected by the voiceprint sensor; a transformer winding voiceprint abnormality judgment module 130, which is used to judge whether the transformer winding has an abnormal voiceprint based on the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector.

[0025] In the above-mentioned transformer abnormal voiceprint diagnosis system 100 based on neural network, the transformer winding data acquisition module 110 is used to obtain the transformer winding thermal image collected by the infrared thermal imager, the transformer winding vibration signal collected by the vibration sensor at multiple predetermined time points, and the transformer winding voiceprint signal collected by the voiceprint sensor at multiple predetermined time points. It should be understood that with the increase of power system capacity and load, the electromagnetic environment of the substation becomes more complex, thereby increasing the risk of transformer failure. The data shows that one of the main causes of transformer failure is the deformation of the winding, and this deformation mainly comes from the unbalanced electromagnetic force generated inside the winding after the external short circuit of the transformer. Although the frequency response analysis method can effectively reflect the internal mechanical state of the transformer, it is highly dependent on professional knowledge, and the diagnostic conclusion is highly subjective, and may even vary from person to person, which may lead to differences in the diagnostic conclusions for the same frequency curve. At the same time, considering that the transformer winding thermal image collected by the infrared thermal imager and the transformer winding vibration signal collected by the vibration sensor at multiple predetermined time points provide some information about the transformer state, they cannot fully reflect the subtle changes and potential problems inside the transformer. Therefore, in the technical solution of the present application, by combining the voiceprint signal to provide unique information about the internal state of the device, the winding voiceprint signal can be combined with the thermal image and the vibration signal to provide more comprehensive equipment monitoring and fault diagnosis. Specifically, the thermal image can show the temperature distribution of the transformer winding. By analyzing the thermal image, the local temperature abnormality of the winding, such as local overheating or overcooling, can be detected, which is helpful to find the fault or abnormality inside the winding. In addition, the thermal image can also be used to monitor the thermal stability and heat distribution uniformity of the winding, which is helpful to evaluate the working state and thermal load of the transformer. The vibration signal can reflect the vibration of the transformer winding. The abnormal vibration signal may indicate that the winding has mechanical problems, such as looseness, wear or resonance. By analyzing the spectral characteristics of the vibration signal, the operating state of the winding can be judged, including whether there is an abnormal vibration frequency or amplitude, which is helpful to predict potential faults. The voiceprint signal can provide sound information inside the winding, which is helpful to detect the sound abnormality inside the winding, such as abnormal noise, discharge sound, etc. The voiceprint signal can also reflect the local fault or abnormality inside the winding. The sound characteristics can be used to judge whether the winding has problems such as discharge and short circuit. The comprehensive use of these three different types of data can provide a more comprehensive understanding of the working status of the transformer windings, improve the accuracy and timeliness of abnormality detection, and help ensure the safe operation and stability of transformer equipment.

[0026] In the above-mentioned transformer abnormal voiceprint diagnosis system 100 based on neural network, the transformer winding data extraction module 120 is used to extract the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector from the transformer winding thermal image acquired by the infrared thermal imager, the transformer winding vibration signal acquired by the vibration sensor at multiple predetermined time points, and the transformer winding voiceprint signal acquired by the voiceprint sensor at multiple predetermined time points. It should be understood that by extracting feature vectors from the thermal image, the thermal characteristics of the winding can be captured, including information such as temperature distribution, thermal stability, and thermal distribution uniformity. The obtained transformer winding thermal image enhancement feature vector can help identify thermal anomalies in the winding, such as local overheating or overcooling areas, so as to timely discover potential faults or problems. The transformer winding global feature vector is a feature vector extracted by integrating multiple information such as vibration signals and voiceprint signals, which can more comprehensively reflect the overall state of the winding. This transformer winding global feature vector can comprehensively consider the vibration characteristics and voiceprint characteristics of the winding, thereby providing a more comprehensive and accurate diagnosis result. By comprehensively analyzing multiple data sources such as thermal images, vibration signals, and voiceprint signals, and extracting the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector, we can have a more comprehensive understanding of the working status of the transformer winding, help predict potential problems and take maintenance measures in a timely manner, thereby improving the reliability and safety of the transformer equipment.

[0027] Figure 2 FIG. 1 is a block diagram of a transformer winding data extraction module in a transformer abnormal voiceprint diagnosis system based on a neural network according to an embodiment of the present application. Figure 2As shown, in a specific embodiment of the present application, the transformer winding data extraction module 120 includes: a transformer winding thermal image feature extraction unit 121, which is used to extract features from the transformer winding thermal image collected by the infrared thermal imager to obtain the transformer winding thermal image enhancement feature vector; a transformer winding vibration feature extraction unit 122, which is used to extract features from the transformer winding vibration signals at multiple predetermined time points collected by the vibration sensor to obtain the transformer winding transformation time-frequency feature vector; a winding voiceprint feature extraction unit 123, which is used to extract features from the transformer winding voiceprint signals at multiple predetermined time points collected by the voiceprint sensor to obtain the transformer winding voiceprint feature vector; a winding feature fusion unit 124, which is used to fuse the transformer winding transformation time-frequency feature vector and the transformer winding voiceprint feature vector to obtain the transformer winding global feature vector. It should be understood that through feature extraction, the original thermal image data can be converted into a more representative and distinguishable feature vector to highlight the key features in the thermal image, thereby improving the sensitivity to the transformer winding condition and further improving the data expression ability. By extracting discriminative feature vectors, the classification and recognition accuracy of transformer winding thermal images can be improved, which helps to quickly and accurately identify the equipment status.

[0028] Furthermore, the vibration signal can reflect the vibration condition of the transformer winding. By extracting the time-frequency features, the frequency and amplitude changes of the vibration signal can be better analyzed, thereby identifying possible signs of faults. It should be understood that time-frequency feature extraction is an effective signal processing method that can combine time domain and frequency domain information to extract the dynamic characteristics of the signal, thereby more comprehensively describing the changing laws of the vibration signal. Combining time-frequency features can better understand the nature of the vibration signal, including the changes in frequency components over time. This helps to conduct a deeper analysis of the vibration signal, thereby improving the understanding of the equipment status.

[0029] Furthermore, the operating status of the transformer winding can be monitored by extracting the voiceprint feature vector. Specifically, the change of the transformer winding voiceprint feature vector can indicate the health status of the equipment and help to detect potential problems early. The process of voiceprint feature vector extraction can help to conduct more in-depth data analysis of the voiceprint signal, so as to understand the characteristics and laws of the transformer winding voiceprint signal and provide support for subsequent decision-making and control.

[0030] In particular, the transformer winding transformation time-frequency feature vector and transformer winding voiceprint feature vector represent the characteristics of the vibration signal and voiceprint signal respectively. Fusion of these two different types of features can provide more comprehensive and integrated information to better describe the characteristics of the transformer winding, thereby better distinguishing the differences between different devices or device states and improving the accuracy of recognition. In addition, fusion of multiple features can enhance the system's resistance to noise and interference and improve the robustness of the system.

[0031] Figure 3 FIG. 1 is a block diagram of a transformer winding thermal image feature extraction unit in a transformer abnormal soundprint diagnosis system based on a neural network according to an embodiment of the present application. Figure 3 As shown, in a specific embodiment of the present application, the transformer winding thermal image feature extraction unit 121 includes: a transformer winding thermal image denoising subunit 1211, which is used to pass the transformer winding thermal image acquired by the infrared thermal imager through a transformer winding thermal image denoising model based on a convolutional neural network to obtain a transformer winding thermal image enhancement image after denoising; a transformer winding thermal image enhancement feature encoding subunit 1212, which is used to feature encode the transformer winding thermal image enhancement image after denoising to obtain the transformer winding thermal image enhancement feature vector. It should be understood that the image acquired by the infrared thermal imager may be affected by noise and interference, and the denoising process helps to highlight the effective features in the image, making the details in the image clearer and more prominent, thereby providing more reliable information and supporting accurate diagnosis and decision-making. Among them, the denoising model based on the convolutional neural network can learn the features in the image, thereby improving the information content and recognition of the image. Specifically, the thermal image of the transformer winding collected by the infrared thermal imager is input into the encoder of the transformer winding thermal image denoising model based on the convolutional neural network, wherein the encoder uses a convolutional layer to perform explicit spatial encoding on the thermal image of the transformer winding collected by the infrared thermal imager to obtain a noise reduction feature; and the noise reduction feature is input into the decoder of the transformer winding thermal image denoising model based on the convolutional neural network, wherein the decoder uses a deconvolution layer to perform deconvolution processing on the noise reduction feature to obtain the transformer winding thermal image enhanced image after noise reduction.

[0032] Furthermore, the feature encoding process can extract the most critical and representative information in the image, which helps reduce the complexity of the data and the storage space requirements. By performing feature encoding on the enhanced image, the important features in the image can be captured, which helps to better describe the thermal imaging characteristics of the transformer winding. These feature vectors contain the key information of the image and can improve the efficiency of subsequent data processing and analysis. At the same time, feature encoding can convert image data into a more compact representation, thereby reducing the computational cost of subsequent processing.

[0033] In a specific embodiment of the present application, the transformer winding thermal image enhancement feature encoding subunit 1212 includes: passing the denoised transformer winding thermal image enhancement image through a transformer winding thermal image enhancement target detection network to obtain a transformer winding thermal image enhancement region of interest; passing the transformer winding thermal image enhancement region of interest through a transformer winding thermal image enhancement feature encoder based on a significant target detector to obtain the transformer winding thermal image enhancement feature vector. It should be understood that the target detection network can improve the accuracy and precision of the region of interest. Through the learning and training of the network, the transformer winding can be detected and located more accurately to avoid missed detection and false detection. Specifically, the target detection network can help locate the target object or region of interest in the image, such as the transformer winding. Through the target detection network, the region of interest in the image can be quickly screened out, reducing the amount of calculation for processing the entire image, thereby improving processing efficiency and speed. More specifically, it is used to use the transformer winding thermal image enhancement target detection network to process the denoised transformer winding thermal image enhancement image with the following target detection formula to obtain the transformer winding thermal image enhancement region of interest;

[0034] Among them, the target detection formula is:

[0035] ROI=H(ψ det ,B)=(cls(ψ det ,B),Regr(ψ det ,B))

[0036] Among them, ψ det is the enhanced image of thermal image of transformer winding after denoising, B is the anchor frame, ROI is the region of interest for thermal image enhancement of transformer winding, cls(ψ det ,B) indicates classification, Regr(ψ det ,B) indicates regression.

[0037] Furthermore, the feature encoder based on the salient object detector can help extract key information and features in the region of interest. This helps capture the most important features in the thermal image of the transformer winding, thereby better describing and understanding the thermal image characteristics of the transformer winding. By combining the region of interest with the salient object for feature encoding, a more representative and discriminative feature vector can be generated to improve the stability and accuracy of feature encoding and avoid the influence of irrelevant information on feature extraction. Specifically, each layer of the transformer winding thermal imaging enhancement feature encoder based on the salient target detector performs the following operations on the input data in the forward pass of the layer: convolution processing is performed on the input data using a first convolution kernel to obtain a first convolution feature map; convolution processing is performed on the first convolution feature map using a second convolution kernel to obtain a second convolution feature map, wherein the size of the first convolution kernel is larger than the size of the second convolution kernel; and, pooling processing is performed on the second convolution feature map to obtain a pooled feature map; and, activation processing is performed on the pooled feature map to obtain an activated feature map; wherein the output of the last layer of the first convolution neural network model is the transformer winding thermal imaging enhancement feature vector, and the input of the first layer of the first convolution neural network model is the transformer winding thermal imaging enhancement region of interest.

[0038] In a specific embodiment of the present application, the transformer winding vibration feature extraction unit 122 includes: performing SIFT transformation on the transformer winding vibration signals at multiple predetermined time points collected by the vibration sensor to obtain a transformer winding SIFT transformation time-frequency diagram; passing the transformer winding SIFT transformation time-frequency diagram through a transformer winding transformation time-frequency feature extractor based on a convolutional neural network to obtain the transformer winding transformation time-frequency feature vector. It should be understood that SIFT transformation is a feature extraction algorithm used in image processing and computer vision, which can extract key points at different scales and rotations to convert signals from the time domain to the frequency domain, and considers features at different scales. Considering that Fourier transform cannot reflect feature information in the time dimension, and when there is an abnormality in a small section on the time axis, the entire Fourier spectrum result will be completely wrong. SIFT is a scale-invariant feature transformation, which uses a windowing method in the time dimension so that the corresponding spectrum corresponds to a specific time period of the signal, and it has very strong robustness. Based on this, the technical solution of the present application performs SIFT transformation on the transformer winding vibration signals at multiple predetermined time points collected by the vibration sensor to obtain a transformer winding SIFT transformation time-frequency diagram.

[0039] Furthermore, a transformer winding transformation time-frequency feature extractor based on a convolutional neural network is used to perform high-dimensional implicit feature mining on the transformer winding SIFT transformation time-frequency graph, so as to extract the feature distribution representation of the local implicit features of the transformer winding SIFT transformation time-frequency graph in a high-dimensional space, thereby obtaining a transformer winding transformation time-frequency feature vector. In this way, the system can be helped to learn a higher-level feature representation, thereby improving the expressiveness and discrimination of the features. It should be understood that the convolutional neural network can effectively capture the time-frequency features in the image when processing the time-frequency image, and help extract the key time-frequency information in the transformer winding vibration signal. This helps to describe the characteristics of the vibration signal more comprehensively. The transformer winding transformation time-frequency feature vector extracted by the convolutional neural network can better represent the characteristics of the vibration signal, thereby improving the accuracy and performance of subsequent classification, recognition or other tasks. Specifically, each layer of the transformer winding transformation time-frequency feature extractor based on the convolutional neural network is used to perform two-dimensional convolution processing, pooling processing along the channel dimension and nonlinear activation processing on the input data in the forward transfer of the layer, so that the last layer of the transformer winding transformation time-frequency feature extractor based on the convolutional neural network outputs the transformer winding transformation time-frequency feature vector, wherein the input of the first layer of the transformer winding transformation time-frequency feature extractor based on the convolutional neural network is the transformer winding SIFT transformation time-frequency map.

[0040] In a specific embodiment of the present application, the winding voiceprint feature extraction unit 123 includes: extracting the logarithmic Mel spectrogram of the transformer winding voiceprint signal at multiple predetermined time points collected by the voiceprint sensor to obtain the transformer winding voiceprint logarithmic Mel spectrogram; passing the transformer winding voiceprint logarithmic Mel spectrogram through a transformer winding voiceprint feature extractor based on a spatial attention mechanism to obtain the transformer winding voiceprint feature vector. It should be understood that the logarithmic Mel spectrogram is one of the commonly used feature representation methods in sound signal processing. By performing Mel spectrum analysis and logarithmic operations on the sound signal, the information of the sound signal in the frequency domain can be extracted, which is helpful for analyzing the energy distribution of different frequency components in the sound signal. The obtained transformer winding voiceprint logarithmic Mel spectrogram can be used as an input feature of the voiceprint recognition system to convert the sound signal into a feature representation form that can be processed by the machine learning algorithm, thereby realizing the recognition and verification of the voiceprint.

[0041] Furthermore, the model based on the spatial attention mechanism can effectively focus on different parts of the input data, thereby improving the model's ability to extract important features. For voiceprint recognition tasks, through the spatial attention mechanism, the model can better focus on the key features in the voiceprint log-Mel spectrogram and improve the distinguishability of features. The spatial attention mechanism helps to introduce interactions and dependencies between features in the feature extraction process, thereby better capturing the complex feature patterns in the transformer winding voiceprint log-Mel spectrogram. This helps to improve the expressiveness of voiceprint features. Specifically, each layer of the transformer winding voiceprint feature extractor based on the spatial attention mechanism performs the following operations on the input data during the forward transfer process of the layer: convolution processing is performed on the input data to generate a convolution feature map; pooling processing is performed on the convolution feature map to generate a pooling feature map; nonlinear activation is performed on the pooling feature map to generate an activation feature map; the mean of each position of the activation feature map along the channel dimension is calculated to generate a spatial feature matrix; the Softmax-like function value of each position in the spatial feature matrix is ​​calculated to obtain a spatial score matrix; and the positional point multiplication of the spatial feature matrix and the spatial score map is calculated to obtain a feature matrix; wherein the last layer of the transformer winding voiceprint feature extractor based on the spatial attention mechanism outputs the transformer winding voiceprint feature vector.

[0042] In the above-mentioned transformer abnormal voiceprint diagnosis system 100 based on neural network, the transformer winding voiceprint abnormality judgment module 130 is used to judge whether the transformer winding has abnormal voiceprint based on the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector. It should be understood that comprehensive consideration of thermal image enhancement features and global features can reduce the possibility of misjudgment. Using a single feature alone may have limitations, while combining multiple features can improve the robustness and reliability of abnormality detection. Among them, the thermal image enhancement feature vector reflects the thermal distribution of the transformer winding, while the global feature vector provides the overall voiceprint feature.

[0043] Figure 4 FIG. 1 is a block diagram of a transformer winding voiceprint abnormality judgment module in a transformer abnormal voiceprint diagnosis system based on a neural network according to an embodiment of the present application. Figure 4As shown, in a specific embodiment of the present application, the transformer winding voiceprint abnormality judgment module 130 includes: a transformer winding feature association unit 131, which is used to associate the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector to obtain a transformer winding abnormal voiceprint diagnosis feature vector; an optimization unit 132, which is used to perform probability density domain dimension derivative correlation projection on the transformer winding abnormal voiceprint diagnosis feature vector to obtain an optimized transformer winding abnormal voiceprint diagnosis feature vector; a winding abnormal voiceprint judgment generation unit 133, which is used to pass the optimized transformer winding abnormal voiceprint diagnosis feature vector through a classifier to obtain a classification result, and the classification result is used to determine whether the transformer winding has an abnormal voiceprint. It should be understood that the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector are complementary. The transformer winding thermal image enhancement feature vector can provide more detailed local information, while the transformer winding global feature vector can provide an overall voiceprint feature. Associating these two features can make up for each other's shortcomings and improve the robustness and accuracy of transformer winding abnormal voiceprint diagnosis.

[0044] In particular, in the technical solution of the present application, it is considered that the transformer winding thermal image enhancement feature vector is a thermal image enhancement feature vector obtained by denoising and extracting the transformer winding thermal image collected by an infrared thermal imager, and the transformer winding global feature vector is a global feature vector extracted by extracting the transformer winding vibration signal collected by a vibration sensor and the transformer winding voiceprint signal collected by a voiceprint sensor. The thermal image, vibration signal and voiceprint signal of the transformer winding all reflect the running state of the winding and contain information about the deformation of the winding. When the transformer winding is deformed, it will not only cause the features in the thermal image to change, but also affect the features of the vibration signal and the voiceprint signal, so there may be a certain correlation between the three signals. In the process of obtaining the transformer winding abnormal voiceprint diagnosis feature vector by fusing the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector, it is considered that there is a correlation between the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector, which causes the probability density function of each position in the feature distribution of the transformer winding abnormal voiceprint diagnosis feature vector to fluctuate abnormally, affecting the robustness of the transformer winding abnormal voiceprint diagnosis feature vector in classification judgment by the classifier. In order to solve this problem, in the technical solution of the present application, the transformer winding abnormal voiceprint diagnosis feature vector is subjected to probability density domain dimension-derived correlation projection to obtain an optimized transformer winding abnormal voiceprint diagnosis feature vector.

[0045] Specifically, the transformer winding abnormal voiceprint diagnosis feature vector is subjected to probability density domain dimension-derived correlation projection using the following correlation projection formula to obtain an optimized transformer winding abnormal voiceprint diagnosis feature vector; wherein the correlation projection formula is:

[0046]

[0047] Among them, V c represents the abnormal soundprint diagnostic feature vector of the transformer winding, represents the calculation of the maximum value of the transformer winding abnormal soundprint diagnosis feature vector, ⊙ represents the position point multiplication, α and β represent weight hyperparameters, and V' represents the optimized transformer winding abnormal soundprint diagnosis feature vector.

[0048] Here, in the technical solution of the present application, the transformer winding abnormal soundprint diagnosis feature vector is subjected to probability density domain dimension derived correlation projection, which utilizes the probability density domain dimension derived idea to map the transformer winding abnormal soundprint diagnosis feature vector from the low-dimensional probability density domain space to the high-dimensional correlation space (nonlinear space guided by the wave function) to aggregate the vector information representation in a quasi-complex function domain. Then, using the correlation projection vector, the feature vector after wave function mapping is projected into a higher-dimensional classification feature space to compensate for the negative impact of the phase offset of each position of the transformer winding abnormal soundprint diagnosis feature vector relative to the vector as a whole on the quasi-probability aggregation (that is, the in-phase enhancement and out-of-phase cancellation based on the wave function principle). In this way, the robustness of the classification judgment of the transformer winding abnormal soundprint diagnosis feature vector is improved.

[0049] Furthermore, the classifier can learn complex patterns and rules based on the training data, thereby improving the accuracy of judgment. It can find hidden features in a large amount of data, so as to more accurately determine whether there are abnormal soundprints in the transformer winding. The classifier can be used to achieve real-time monitoring of the transformer winding status, help to promptly discover and handle abnormal situations, and ensure the safety and stability of equipment operation.

[0050] In summary, the embodiment of the present application first obtains the thermal image of the transformer winding collected by the infrared thermal imager, the transformer winding vibration signal at multiple predetermined time points collected by the vibration sensor, and the transformer winding voiceprint signal at multiple predetermined time points collected by the voiceprint sensor, and then uses deep learning technology to perform feature extraction and correlation analysis on the three, and finally obtains the classification result through the classifier to determine whether the transformer winding has abnormal voiceprints, thereby determining whether the transformer winding has deformation, which helps to detect problems with the transformer winding as early as possible, and take corresponding repair and maintenance measures to avoid further damage and accidents, and further achieve preventive maintenance and improve power grid reliability.

[0051] As described above, the transformer abnormal voiceprint diagnosis system 100 based on neural network according to the embodiment of the present application can be implemented in various terminal devices, such as a server deployed with a transformer abnormal voiceprint diagnosis algorithm based on neural network. In one example, the transformer abnormal voiceprint diagnosis system 100 based on neural network can be integrated into the terminal device as a software module and / or hardware module. For example, the transformer abnormal voiceprint diagnosis system 100 based on neural network can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the transformer abnormal voiceprint diagnosis system 100 based on neural network can also be one of the many hardware modules of the terminal device.

[0052] Alternatively, in another example, the neural network-based transformer abnormal soundprint diagnosis system 100 and the terminal device may also be separate devices, and the neural network-based transformer abnormal soundprint diagnosis system 100 may be connected to the terminal device via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0053] Exemplary Methods

[0054] Figure 5 FIG. 1 is a flowchart of a transformer abnormal soundprint diagnosis method based on a neural network according to an embodiment of the present application. Figure 5 As shown, according to the neural network-based transformer abnormal soundprint diagnosis method of the embodiment of the present application, it includes: S110, obtaining the transformer winding thermal image collected by the infrared thermal imager, the transformer winding vibration signal at multiple predetermined time points collected by the vibration sensor, and the transformer winding soundprint signal at multiple predetermined time points collected by the soundprint sensor; S120, extracting the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector from the transformer winding thermal image collected by the infrared thermal imager, the transformer winding vibration signal at multiple predetermined time points collected by the vibration sensor, and the transformer winding soundprint signal at multiple predetermined time points collected by the soundprint sensor; S130, judging whether the transformer winding has an abnormal soundprint based on the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector.

[0055] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned transformer abnormal soundprint diagnosis method based on neural network have been referred to above. Figures 1 to 4 The description of the neural network-based transformer abnormal voiceprint diagnosis system has been introduced in detail, and therefore, its repeated description will be omitted.

[0056] Exemplary Electronic Devices

[0057] Below, reference Figure 6 To describe an electronic device according to an embodiment of the present application.

[0058] like Figure 6 As shown, the electronic device 10 includes an input device 11, an input interface 12, a central processing unit 13, a memory 14, an output interface 15, an output device 16, and a bus 17. The input interface 12, the central processing unit 13, the memory 14, and the output interface 15 are connected to each other through the bus 17, and the input device 11 and the output device 16 are connected to the bus 17 through the input interface 12 and the output interface 15, respectively, and then connected to other components of the electronic device 10.

[0059] Specifically, the input device 11 receives input information from the outside and transmits the input information to the central processing unit 13 through the input interface 12; the central processing unit 13 processes the input information based on the computer executable instructions stored in the memory 14 to generate output information, temporarily or permanently stores the output information in the memory 14, and then transmits the output information to the output device 16 through the output interface 15; the output device 16 outputs the output information to the outside of the electronic device 10 for user use.

[0060] In one embodiment, Figure 6 The electronic device 10 shown can be implemented as a network device, which may include: a memory configured to store programs; a processor configured to run the programs stored in the memory to execute any one of the neural network-based transformer abnormal soundprint diagnosis methods described in the above embodiments.

[0061] According to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program tangibly contained on a machine-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network, and / or installed from a removable storage medium.

[0062] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0063] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present application, but the present application is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and substance of the present application, and these modifications and improvements are also considered to be within the scope of protection of the present application.

Claims

1. A transformer abnormal voiceprint diagnosis system based on neural network, characterized in that: include: A transformer winding data acquisition module, used to acquire a transformer winding thermal image acquired by an infrared thermal imager, transformer winding vibration signals acquired by a vibration sensor at multiple predetermined time points, and transformer winding voiceprint signals acquired by a voiceprint sensor at multiple predetermined time points; A transformer winding data extraction module, used to extract a transformer winding thermal image enhancement feature vector and a transformer winding global feature vector from the transformer winding thermal image acquired by the infrared thermal imager, the transformer winding vibration signals at multiple predetermined time points acquired by the vibration sensor, and the transformer winding voiceprint signals at multiple predetermined time points acquired by the voiceprint sensor; The transformer winding voiceprint abnormality judgment module is used to judge whether the transformer winding has abnormal voiceprint based on the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector.

2. The transformer abnormal voiceprint diagnosis system based on neural network according to claim 1 is characterized in that: The transformer winding data extraction module comprises: A transformer winding thermal image feature extraction unit, used for extracting features from the transformer winding thermal image acquired by the infrared thermal imager to obtain the transformer winding thermal image enhancement feature vector; A transformer winding vibration feature extraction unit, used for extracting features from the transformer winding vibration signals at a plurality of predetermined time points collected by the vibration sensor to obtain a transformer winding transformation time-frequency feature vector; A winding voiceprint feature extraction unit, used for extracting features of transformer winding voiceprint signals collected by the voiceprint sensor at a plurality of predetermined time points to obtain a transformer winding voiceprint feature vector; The winding feature fusion unit is used to fuse the transformer winding transformation time-frequency feature vector and the transformer winding voiceprint feature vector to obtain the transformer winding global feature vector.

3. The transformer abnormal voiceprint diagnosis system based on neural network according to claim 2 is characterized in that: The transformer winding thermal image feature extraction unit comprises: The transformer winding thermal image denoising subunit is used to pass the transformer winding thermal image acquired by the infrared thermal imager through a transformer winding thermal image denoising model based on a convolutional neural network to obtain a transformer winding thermal image enhanced image after denoising; The transformer winding thermal image enhancement feature encoding subunit is used to perform feature encoding on the denoised transformer winding thermal image enhancement image to obtain the transformer winding thermal image enhancement feature vector.

4. The transformer abnormal voiceprint diagnosis system based on neural network according to claim 3 is characterized in that: The transformer winding thermal image enhancement feature encoding subunit comprises: Passing the de-noised transformer winding thermal image enhancement image through a transformer winding thermal image enhancement target detection network to obtain a transformer winding thermal image enhancement region of interest; The transformer winding thermal image enhancement region of interest is passed through a transformer winding thermal image enhancement feature encoder based on a significant target detector to obtain the transformer winding thermal image enhancement feature vector.

5. The transformer abnormal voiceprint diagnosis system based on neural network according to claim 4 is characterized in that: The transformer winding vibration feature extraction unit comprises: Performing SIFT transformation on the transformer winding vibration signals at a plurality of predetermined time points collected by the vibration sensor to obtain a transformer winding SIFT transformation time-frequency diagram; The transformer winding SIFT transformed time-frequency diagram is passed through a transformer winding transformation time-frequency feature extractor based on a convolutional neural network to obtain the transformer winding transformation time-frequency feature vector.

6. The transformer abnormal voiceprint diagnosis system based on neural network according to claim 5 is characterized in that: The winding voiceprint feature extraction unit comprises: Extracting the logarithmic Mel-spectrogram of the transformer winding voiceprint signal collected by the voiceprint sensor at a plurality of predetermined time points to obtain the logarithmic Mel-spectrogram of the transformer winding voiceprint; The transformer winding voiceprint logarithmic Mel-spectrogram is passed through a transformer winding voiceprint feature extractor based on a spatial attention mechanism to obtain the transformer winding voiceprint feature vector.

7. The transformer abnormal voiceprint diagnosis system based on neural network according to claim 6 is characterized in that: The transformer winding voiceprint abnormality judgment module includes: A transformer winding feature association unit, used for associating the transformer winding thermal image enhancement feature vector with the transformer winding global feature vector to obtain a transformer winding abnormal voiceprint diagnosis feature vector; An optimization unit, used for performing probability density domain dimension-derived correlation projection on the transformer winding abnormal voiceprint diagnosis feature vector to obtain an optimized transformer winding abnormal voiceprint diagnosis feature vector; The winding abnormal voiceprint judgment generating unit is used to pass the optimized transformer winding abnormal voiceprint diagnosis feature vector through a classifier to obtain a classification result, and the classification result is used to judge whether the transformer winding has an abnormal voiceprint.

8. The transformer abnormal voiceprint diagnosis system based on neural network according to claim 7 is characterized in that: The optimization unit comprises: The transformer winding abnormal voiceprint diagnosis feature vector is subjected to probability density domain dimension-derived correlation projection using the following correlation projection formula to obtain an optimized transformer winding abnormal voiceprint diagnosis feature vector; Wherein, the correlation projection formula is: Among them, V c represents the abnormal soundprint diagnostic feature vector of the transformer winding, represents the calculation of the maximum value of the transformer winding abnormal soundprint diagnosis feature vector, ⊙ represents the position point multiplication, α and β represent weight hyperparameters, and V' represents the optimized transformer winding abnormal soundprint diagnosis feature vector.

9. A transformer abnormal soundprint diagnosis method based on neural network, characterized in that: include: Acquire a thermal image of a transformer winding acquired by an infrared thermal imager, a vibration signal of the transformer winding acquired by a vibration sensor at a plurality of predetermined time points, and a voiceprint signal of the transformer winding acquired by a voiceprint sensor at a plurality of predetermined time points; Extracting a transformer winding thermal image enhancement feature vector and a transformer winding global feature vector from the transformer winding thermal image acquired by the infrared thermal imager, the transformer winding vibration signals at multiple predetermined time points acquired by the vibration sensor, and the transformer winding voiceprint signals at multiple predetermined time points acquired by the voiceprint sensor; Based on the transformer winding thermal image enhancement feature vector and the transformer winding global feature vector, it is determined whether the transformer winding has an abnormal soundprint.

10. The transformer abnormal soundprint diagnosis method based on neural network according to claim 9 is characterized in that: Extracting a transformer winding thermal image enhancement feature vector and a transformer winding global feature vector from the transformer winding thermal image acquired by the infrared thermal imager, the transformer winding vibration signals at multiple predetermined time points acquired by the vibration sensor, and the transformer winding voiceprint signals at multiple predetermined time points acquired by the voiceprint sensor, including: Performing feature extraction on the transformer winding thermal image acquired by the infrared thermal imager to obtain the transformer winding thermal image enhancement feature vector; Performing feature extraction on the transformer winding vibration signals collected by the vibration sensor at a plurality of predetermined time points to obtain a transformer winding transformation time-frequency feature vector; Performing feature extraction on the transformer winding voiceprint signals collected by the voiceprint sensor at a plurality of predetermined time points to obtain a transformer winding voiceprint feature vector; The transformer winding transformation time-frequency feature vector and the transformer winding voiceprint feature vector are fused to obtain the transformer winding global feature vector.

Images