Transformer defect detection system and method based on voiceprint detection

Abstract

The embodiment of the invention provides a transformer defect detection system based on voiceprint detection and a method thereof. The method comprises the following steps: firstly, acquiring a sound wave signal of a detected transformer collected by a sound wave sensor, then, carrying out time-frequency analysis on the sound wave signal to obtain voiceprint characteristic parameters, then, carrying out local waveform characteristic extraction on the sound wave signal to obtain a sequence of sound wave fragment signal waveform characteristic vectors, and finally, carrying out local waveform characteristic extraction on the sequence of the sound wave fragment signal waveform characteristic vectors. The voiceprint characteristic parameters are integrated into a sequence of the sound wave fragment signal waveform characteristic vectors to obtain a voiceprint characteristic embedded guiding sound wave signal global waveform characteristic vector, and finally, whether the detected transformer has a fault or not is determined based on the voiceprint characteristic embedded guiding sound wave signal global waveform characteristic vector. Therefore, transformer fault detection and identification can be realized.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of transformer defect detection, and more specifically, to a transformer defect detection system and method based on voiceprint detection. Background Art

[0002] A transformer is an important device in the power system, and its operating state directly affects the safety and stability of the power system. Faults in transformers often lead to serious consequences such as power supply interruptions, equipment damage, and even fires. Therefore, timely and effective detection of transformer faults and defects is an important measure to ensure the normal operation of the power system.

[0003] Traditional transformer defect detection relies on manual operation, which has some problems. For example, it has a relatively high invasiveness to equipment and requires relatively high professional knowledge. Therefore, an optimized transformer defect detection scheme is expected. Summary of the Invention

[0004] To solve the above technical problems, the present invention is proposed. The embodiments of the present invention provide a transformer defect detection system and method based on voiceprint detection, which use voiceprint detection and deep learning algorithms to process and analyze the acoustic wave signals of the transformer to be detected, and can achieve non-invasive transformer fault detection and identification.

[0005] In a first aspect, the embodiments of the present invention provide a transformer defect detection method based on voiceprint detection, which includes:

[0006] Obtaining the acoustic wave signals of the transformer to be detected collected by an acoustic wave sensor;

[0007] Performing time-frequency analysis on the acoustic wave signals to obtain voiceprint feature parameters;

[0008] Extracting local waveform features from the acoustic wave signals to obtain a sequence of acoustic wave segment signal waveform feature vectors;

[0009] Integrating the voiceprint feature parameters into the sequence of acoustic wave segment signal waveform feature vectors to obtain a voiceprint feature embedded guided acoustic wave signal global waveform feature vector; and

[0010] Determining whether there is a fault in the transformer to be detected based on the voiceprint feature embedded guided acoustic wave signal global waveform feature vector.

[0011] In some possible embodiments, performing time-frequency analysis on the acoustic wave signals to obtain voiceprint feature parameters includes: performing time-frequency analysis based on short-time Fourier transform on the acoustic wave signals to obtain the voiceprint feature parameters.

[0012] In some possible embodiments, local waveform feature extraction is performed on the acoustic wave signal to obtain a sequence of acoustic wave segment signal waveform feature vectors, including:

[0013] Signal segmentation is performed on the acoustic wave signal to obtain a sequence of acoustic wave segment signals; and

[0014] The sequence of acoustic wave segment signals is respectively passed through an acoustic wave waveform feature extractor based on a convolutional neural network model to obtain the sequence of acoustic wave segment signal waveform feature vectors.

[0015] In some possible embodiments, the acoustic wave waveform feature extractor based on a convolutional neural network model includes an input layer, a convolutional layer, an activation layer, a pooling layer, and an output layer.

[0016] In some possible embodiments, passing the sequence of acoustic wave segment signals through the acoustic wave waveform feature extractor based on a convolutional neural network model to obtain the sequence of acoustic wave segment signal waveform feature vectors includes:

[0017] Using each layer of the acoustic wave waveform feature extractor based on a convolutional neural network model to perform two-dimensional convolutional processing, mean pooling processing based on a feature matrix, and non-linear activation processing on the input data respectively in the forward pass of the layer, and outputting the sequence of acoustic wave segment signal waveform feature vectors by the last layer of the acoustic wave waveform feature extractor based on a convolutional neural network model, wherein the input of the first layer of the acoustic wave waveform feature extractor based on a convolutional neural network model is the sequence of acoustic wave segment signals.

[0018] In some possible embodiments, incorporating the voiceprint feature parameters into the sequence of acoustic wave segment signal waveform feature vectors to obtain a voiceprint feature-embedded guided acoustic wave signal global waveform feature vector includes:

[0019] Arranging the voiceprint feature parameters into a voiceprint feature parameter input vector; and

[0020] Inputting the sequence of acoustic wave segment signal waveform feature vectors and the voiceprint feature parameter input vector into a feature embedding module to obtain the voiceprint feature-embedded guided acoustic wave signal global waveform feature vector.

[0021] In some possible embodiments, inputting the sequence of acoustic wave segment signal waveform feature vectors and the voiceprint feature parameter input vector into a feature embedding module to obtain the voiceprint feature-embedded guided acoustic wave signal global waveform feature vector includes:

[0022] Passing the voiceprint feature parameter input vector through a feature extractor based on a fully convolutional network model to obtain a voiceprint feature temporal feature vector;

[0023] Linearly process the acoustic feature time-series feature vector to obtain a linearly processed acoustic feature time-series feature vector;

[0024] Linearly process the sequence of the acoustic wave segment signal waveform feature vectors to obtain a sequence of linearly processed acoustic wave segment signal waveform feature vectors;

[0025] Fuse the linearly processed acoustic feature time-series feature vector and the sequence of the linearly processed acoustic wave segment signal waveform feature vectors to obtain an acoustic-wave linear initial fusion vector;

[0026] Perform one-dimensional convolution processing on the sequence of the acoustic wave segment signal waveform feature vectors to obtain a sequence of acoustic time-series neighborhood correlation feature vectors; and

[0027] Fuse the sequence of the acoustic time-series neighborhood correlation feature vectors and the acoustic-wave linear initial fusion vector in a splicing manner to obtain the acoustic feature embedding-guided acoustic wave signal global waveform feature vector.

[0028] In some possible embodiments, based on the acoustic feature embedding-guided acoustic wave signal global waveform feature vector, determining whether the detected transformer has a fault includes:

[0029] Perform feature distribution optimization on the acoustic feature embedding-guided acoustic wave signal global waveform feature vector to obtain an optimized acoustic feature embedding-guided acoustic wave signal global waveform feature vector; and

[0030] Pass the optimized acoustic feature embedding-guided acoustic wave signal global waveform feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the detected transformer has a fault.

[0031] In a second aspect, an embodiment of the present invention provides a transformer defect detection system based on acoustic fingerprint detection, which includes:

[0032] An acoustic wave signal acquisition module, configured to acquire an acoustic wave signal of a detected transformer collected by an acoustic wave sensor;

[0033] A time-frequency analysis module, configured to perform time-frequency analysis on the acoustic wave signal to obtain acoustic feature parameters;

[0034] A local waveform feature extraction module, configured to extract local waveform features of the acoustic wave signal to obtain a sequence of acoustic wave segment signal waveform feature vectors;

[0035] An acoustic feature integration module, configured to integrate the acoustic feature parameters into the sequence of the acoustic wave segment signal waveform feature vectors to obtain an acoustic feature embedding-guided acoustic wave signal global waveform feature vector; and

[0036] A fault analysis module, configured to determine whether there is a fault in the transformer under test based on the acoustic feature embedding-guided global waveform feature vector of the acoustic wave signal.

[0037] In some possible embodiments, the time-frequency analysis module is configured to: perform time-frequency analysis on the acoustic wave signal based on short-time Fourier transform to obtain the acoustic feature parameters.

[0038] Compared with the prior art, the transformer defect detection system and method based on acoustic fingerprint detection provided by the present invention first acquires the acoustic wave signal of the transformer under test collected by an acoustic wave sensor, then performs time-frequency analysis on the acoustic wave signal to obtain acoustic feature parameters, then extracts local waveform features of the acoustic wave signal to obtain a sequence of acoustic wave segment signal waveform feature vectors, then incorporates the acoustic feature parameters into the sequence of acoustic wave segment signal waveform feature vectors to obtain an acoustic feature embedding-guided global waveform feature vector of the acoustic wave signal, and finally determines whether there is a fault in the transformer under test based on the acoustic feature embedding-guided global waveform feature vector of the acoustic wave signal. In this way, non-invasive transformer fault detection and identification can be achieved. Description of the Drawings

[0039] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0040] Figure 1 It is a flowchart of a transformer defect detection method based on acoustic fingerprint detection according to an embodiment of the present invention;

[0041] Figure 2 It is a schematic structural diagram of a transformer defect detection method based on acoustic fingerprint detection according to an embodiment of the present invention;

[0042] Figure 3 It is a flowchart of sub-step S130 of a transformer defect detection method based on acoustic fingerprint detection according to an embodiment of the present invention;

[0043] Figure 4 It is a flowchart of sub-step S140 of a transformer defect detection method based on acoustic fingerprint detection according to an embodiment of the present invention;

[0044] Figure 5 It is a flowchart of sub-step S150 of a transformer defect detection method based on acoustic fingerprint detection according to an embodiment of the present invention;

[0045] Figure 6Block diagram of a transformer defect detection system based on voiceprint detection according to an embodiment of the present invention;

[0046] Figure 7 Application scenario diagram of a transformer defect detection method based on voiceprint detection according to an embodiment of the present invention. Detailed implementation manners

[0047] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation manners. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0048] Unless otherwise specifically stated, the technical terms or scientific terms used in the embodiments of the present invention shall have the ordinary meaning understood by those of ordinary skill in the art to which the present invention pertains. The use of "including" or "comprising" etc. in the embodiments of the present invention does not limit the shapes, numbers, steps, actions, operations, components, elements and / or their groups mentioned, nor does it exclude the appearance or addition of one or more other different shapes, numbers, steps, actions, operations, components, elements and / or their groups, or the addition of these. In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity and order of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of the present invention, "a plurality of" means two or more, unless otherwise specifically and clearly defined.

[0049] Unless otherwise specifically stated, the relative settings, numerical expressions and values of the components and steps described in these embodiments do not limit the scope of the present invention. At the same time, it should be understood that, for the sake of convenience of description, the dimensions of the various parts shown in the drawings are not drawn in actual proportional relationships. For technologies, methods and devices known to those of ordinary skill in the relevant fields, they may not be discussed in detail, but in appropriate cases, the shown technologies, methods and devices should be regarded as part of the authorization specification. In all the examples shown and discussed here, any specific other example may have different values. It should be noted that: similar symbols and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0050] In the description of the embodiments of the present invention, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In the embodiments of the present invention, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in the embodiments of the present invention and the features of different embodiments or examples.

[0051] Next, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments of the present invention. It should be understood that the present invention is not limited by the exemplary embodiments described herein.

[0052] In view of the above technical problems, the technical concept of the present invention is to use voiceprint detection and deep learning algorithms to process and analyze the acoustic wave signals of the transformer to be detected, so as to achieve non-invasive transformer fault detection and identification.

[0053] Based on this, Figure 1 FIG. is a flowchart of a transformer defect detection method based on voiceprint detection according to an embodiment of the present invention. Figure 2 FIG. is a schematic structural diagram of a transformer defect detection method based on voiceprint detection according to an embodiment of the present invention. As Figure 1 and Figure 2 shown, the transformer defect detection method based on voiceprint detection according to an embodiment of the present invention includes the steps of: S110, acquiring the acoustic wave signal of the transformer to be detected collected by an acoustic wave sensor; S120, performing time-frequency analysis on the acoustic wave signal to obtain voiceprint feature parameters; S130, extracting local waveform features of the acoustic wave signal to obtain a sequence of acoustic wave segment signal waveform feature vectors; S140, integrating the voiceprint feature parameters into the sequence of acoustic wave segment signal waveform feature vectors to obtain a voiceprint feature-embedded guided acoustic wave signal global waveform feature vector; and, S150, determining whether there is a fault in the transformer to be detected based on the voiceprint feature-embedded guided acoustic wave signal global waveform feature vector.

[0054] It should be understood that in step S110, an acoustic wave sensor is used to sample the transformer to be detected, and an acoustic wave signal generated by the transformer is obtained. In step S120, by performing time-frequency analysis on the acoustic wave signal, voiceprint feature parameters can be extracted. Time-frequency analysis is a method of analyzing a signal in terms of time and frequency, which can reveal the variation of the frequency components of the signal over time. In step S130, local waveform feature extraction is performed on the acoustic wave signal. Local waveform feature extraction is a method of extracting local features from the acoustic wave signal, which can capture the detailed information in the acoustic wave signal. The extracted features are organized into a sequence, and each feature vector represents the waveform features of an acoustic wave segment. In step S140, the voiceprint feature parameters are incorporated into the sequence of waveform feature vectors of the acoustic wave segment signals. By doing so, the voiceprint features can be combined with the waveform features of the acoustic wave segment to form a voiceprint feature-embedded guided global waveform feature vector of the acoustic wave signal. This global waveform feature vector contains both the voiceprint features and the waveform features of the acoustic wave signal. Step S150 is to determine whether the transformer to be detected has a fault based on the voiceprint feature-embedded guided global waveform feature vector of the acoustic wave signal. By analyzing and comparing the voiceprint feature-embedded guided acoustic wave signal, it can be judged whether the transformer has defects or faults. This method combines the acoustic wave signal and the voiceprint features, and detects the defects of the transformer by analyzing the time-frequency features, local waveform features, and voiceprint features of the acoustic wave signal. This method can provide a non-contact means for detecting transformer defects, and has a certain degree of practicality and accuracy.

[0055] Specifically, in the technical solution of the present invention, first, an acoustic wave signal of the transformer to be detected collected by an acoustic wave sensor is obtained. It should be understood that during the operation of the transformer, acoustic wave signals of different types and intensities are generated, and these acoustic wave signals reflect the physical state and working conditions inside the transformer. When a transformer fails or has a defect, such as damage to the insulating material, internal discharge, etc., it will cause abnormal changes in the acoustic wave signal, such as changes in parameters such as frequency, amplitude, and phase. By analyzing and processing the acoustic wave signal generated during the operation of the transformer, feature information related to the fault can be extracted, and this feature information can be used as a basis for judging whether the transformer has a fault.

[0056] Then, time-frequency analysis is performed on the acoustic wave signal to obtain voiceprint feature parameters. In a specific example of the present invention, the implementation manner of performing time-frequency analysis on the acoustic wave signal to obtain voiceprint feature parameters is to perform time-frequency analysis based on the short-time Fourier transform (STFT) on the acoustic wave signal to obtain voiceprint feature parameters.

[0057] Here, the short-time Fourier transform can decompose the acoustic wave signal into a series of small segments in the time domain and frequency domain. By calculating the energy or power spectral density of each segment at different frequencies, acoustic fingerprint feature parameters can be obtained. These parameters can reflect the energy distribution of the acoustic wave signal at different frequencies, thereby revealing the operating state of the transformer.

[0058] Correspondingly, in step S120, performing time-frequency analysis on the acoustic wave signal to obtain acoustic fingerprint feature parameters includes: performing time-frequency analysis based on the short-time Fourier transform on the acoustic wave signal to obtain the acoustic fingerprint feature parameters. It should be understood that the short-time Fourier transform (STFT) is a time-frequency analysis method used to analyze signals in the time and frequency domains. It divides the signal into a series of short-time windows and applies the Fourier transform to each window to obtain the spectral information of the signal within that window. The main purpose of the short-time Fourier transform is to establish a connection between the time domain and the frequency domain to better understand the time-varying frequency characteristics of the signal. It can provide the spectral information of the signal at different time periods and reveal the change of the frequency components of the signal over time. By performing short-time Fourier transform analysis on the acoustic wave signal, acoustic fingerprint feature parameters can be obtained. Using the short-time Fourier transform for time-frequency analysis is that acoustic fingerprint feature extraction can convert the sound signal into spectral information and then extract the acoustic fingerprint feature parameters.

[0059] Next, extracting local waveform features of the acoustic wave signal to obtain a sequence of acoustic wave segment signal waveform feature vectors; and integrating the acoustic fingerprint feature parameters into the sequence of acoustic wave segment signal waveform feature vectors to obtain an acoustic fingerprint feature-embedded guided acoustic wave signal global waveform feature vector. Among them, the acoustic wave segment signal waveform feature vector can characterize the local waveform features of the acoustic wave signal and can capture the transient changes caused by faults. The acoustic fingerprint feature parameters can reflect the energy distribution of the acoustic wave signal and can reveal the real-time state of the transformer during operation. Here, fusing the two can fully express the complexity and diversity of the acoustic wave signal.

[0060] Specifically, in an embodiment of the present invention, the sound wave signal is first segmented to obtain a sequence of sound wave segment signals; then, the sequence of sound wave segment signals is respectively passed through a sound wave waveform feature extractor based on a convolutional neural network model to obtain a sequence of sound wave segment signal waveform feature vectors; and then, the sequence of sound wave segment signal waveform feature vectors and the voiceprint feature parameter input vector obtained by arranging the voiceprint feature parameters are input into a feature embedding module to obtain a voiceprint feature embedding-guided sound wave signal global waveform feature vector. Among them, the feature embedding module uses the voiceprint feature information expressed by the voiceprint feature parameter input vector as inducing information to optimize and guide the sound wave waveform feature distribution of each sound wave segment signal waveform feature vector, and then integrates the global information to obtain the voiceprint feature embedding-guided sound wave signal global waveform feature vector that integrates the voiceprint feature information and the global waveform feature information.

[0061] Accordingly, if Figure 3 As shown, in step S130, local waveform feature extraction is performed on the sound wave signal to obtain a sequence of waveform feature vectors of sound wave segment signals, including: S131, signal segmentation is performed on the sound wave signal to obtain a sequence of sound wave segment signals; and, S132, the sequence of sound wave segment signals is respectively passed through a sound wave waveform feature extractor based on a convolutional neural network model to obtain a sequence of waveform feature vectors of the sound wave segment signals.

[0062] It should be understood that the effect of step S131 is to perform signal segmentation on the sound wave signal, and the entire sound wave signal is divided into multiple segment signals. The purpose of doing so is to decompose the long sound wave signal into short segments, so as to better capture the local characteristics of the sound wave signal. By segmenting the sound wave signal, the sound wave signal can be divided into multiple time windows, and the signal in each window is considered to be a relatively independent segment. The effect of step S132 is to process the segmented sound wave segment signal through a sound wave waveform feature extractor based on a convolutional neural network model to obtain a waveform feature vector of each sound wave segment signal. In general, step S131 divides the sound wave signal into multiple segments, and step S132 uses a convolutional neural network model to extract the waveform feature vector of each segment. The purpose of these two steps is to convert the sound wave signal into a feature representation that can be used for further processing and analysis, so as to perform sound-related tasks.

[0063] It is worth mentioning that the Convolutional Neural Network (CNN) is a deep learning model that can effectively learn and extract local features of signals. By inputting the acoustic wave segment signals into the convolutional neural network, the network can automatically learn and extract the waveform features of each segment signal, such as spectral shape, time-domain characteristics, etc. The core idea of the convolutional neural network is to extract local features of the input data through convolutional operations and abstract and combine features in a stacked manner. It consists of multiple layers, including convolutional layers, pooling layers, and fully connected layers. In the convolutional layer, by defining a set of convolutional kernels (also called filters), convolutional operations are performed on the input data to extract local features of the input data. Convolutional operations can capture the spatial structure information of the input data, such as features like edges and textures in images. The pooling layer is used to reduce the size of the feature map and retain important features. The most commonly used pooling operation is max pooling, which selects the maximum value in each region as the pooling result. The fully connected layer flattens the features extracted in the previous layers and performs tasks such as classification or regression through fully connected neurons. By stacking multiple convolutional layers, pooling layers, and fully connected layers in the convolutional neural network, the network can learn higher-level abstract features, thereby achieving more accurate classification or prediction of the input data. In acoustic signal processing, the convolutional neural network can be used as an acoustic wave feature extractor to extract useful information by learning the local features of acoustic wave signals.

[0064] Among them, in step S132, the acoustic wave waveform feature extractor based on the convolutional neural network model includes an input layer, a convolutional layer, an activation layer, a pooling layer, and an output layer. Specifically, passing the sequences of the acoustic wave segment signals through the acoustic wave waveform feature extractor based on the convolutional neural network model respectively to obtain the sequences of the acoustic wave segment signal waveform feature vectors includes: using each layer of the acoustic wave waveform feature extractor based on the convolutional neural network model to perform two-dimensional convolutional processing, mean pooling processing based on the feature matrix, and non-linear activation processing on the input data respectively in the forward pass of the layer to output the sequences of the acoustic wave segment signal waveform feature vectors by the last layer of the acoustic wave waveform feature extractor based on the convolutional neural network model, where the input of the first layer of the acoustic wave waveform feature extractor based on the convolutional neural network model is the sequences of the acoustic wave segment signals.

[0065] Correspondingly, as Figure 4As shown, in step S140, incorporating the voiceprint feature parameters into the sequence of the waveform feature vectors of the acoustic wave segment signals to obtain the voiceprint feature-embedded guided global waveform feature vectors of the acoustic wave signals includes: S141, arranging the voiceprint feature parameters into a voiceprint feature parameter input vector; and S142, inputting the sequence of the waveform feature vectors of the acoustic wave segment signals and the voiceprint feature parameter input vector into a feature embedding module to obtain the voiceprint feature-embedded guided global waveform feature vectors of the acoustic wave signals.

[0066] It should be understood that the function of step S141 is to arrange the voiceprint feature parameters into a voiceprint feature parameter input vector. The voiceprint feature parameters are parameters related to the voice characteristics extracted from the acoustic wave signals, such as spectral characteristics, formants, etc. Arranging these parameters into a vector form is for the convenience of subsequent processing and fusion. The function of step S142 is to input the sequence of the waveform feature vectors of the acoustic wave segment signals and the voiceprint feature parameter input vector into a feature embedding module to obtain the voiceprint feature-embedded guided global waveform feature vectors of the acoustic wave signals. The feature embedding module fuses the voiceprint feature parameters with the waveform features of the acoustic wave segment signals to generate a global waveform feature vector. This global feature vector contains the voiceprint features and the waveform features of the acoustic wave signals, and is used for further analysis and judgment of whether the transformer has faults. By embedding the voiceprint feature parameters into the global waveform feature vectors of the acoustic wave signals, the frequency domain and time domain information of the sound can be comprehensively utilized to improve the accuracy and robustness of transformer defect detection.

[0067] Among them, in step S142, inputting the sequence of the waveform feature vectors of the acoustic wave segment signals and the voiceprint feature parameter input vector into a feature embedding module to obtain the voiceprint feature-embedded guided global waveform feature vectors of the acoustic wave signals includes: passing the voiceprint feature parameter input vector through a feature extractor based on a fully convolutional network model to obtain a voiceprint feature temporal feature vector; linearly processing the voiceprint feature temporal feature vector to obtain a linearly processed voiceprint feature temporal feature vector; linearly processing the sequence of the waveform feature vectors of the acoustic wave segment signals to obtain a sequence of linearly processed waveform feature vectors of the acoustic wave segment signals; fusing the linearly processed voiceprint feature temporal feature vector and the sequence of the linearly processed waveform feature vectors of the acoustic wave segment signals to obtain a voiceprint-acoustic wave linear initial fusion vector; performing one-dimensional convolutional processing on the sequence of the waveform feature vectors of the acoustic wave segment signals to obtain a sequence of acoustic wave temporal neighborhood correlation feature vectors; and fusing the sequence of the acoustic wave temporal neighborhood correlation feature vectors and the voiceprint-acoustic wave linear initial fusion vector in a splicing manner to obtain the voiceprint feature-embedded guided global waveform feature vectors of the acoustic wave signals.

[0068] Subsequently, embed the voiceprint feature into the global waveform feature vector of the guided acoustic wave signal and pass it through a classifier to obtain a classification result, where the classification result is used to indicate whether the detected transformer has a fault.

[0069] Correspondingly, as Figure 5 shown, in step S150, based on embedding the voiceprint feature into the global waveform feature vector of the guided acoustic wave signal, determining whether the detected transformer has a fault includes: S151, optimizing the feature distribution of the voiceprint feature embedded in the global waveform feature vector of the guided acoustic wave signal to obtain an optimized voiceprint feature embedded in the global waveform feature vector of the guided acoustic wave signal; and, S152, passing the optimized voiceprint feature embedded in the global waveform feature vector of the guided acoustic wave signal through a classifier to obtain a classification result, where the classification result is used to indicate whether the detected transformer has a fault.

[0070] It should be understood that the function of step S151 is to optimize the feature distribution of the voiceprint feature embedded in the global waveform feature vector of the guided acoustic wave signal to obtain an optimized voiceprint feature embedded in the global waveform feature vector of the guided acoustic wave signal. In this step, the global waveform feature vector of the voiceprint feature embedded in the guided acoustic wave signal is optimized. The purpose is to improve the distribution of the features so that the features are more conducive to subsequent classification tasks. Through feature distribution optimization, the accuracy of transformer fault detection can be further improved. The function of step S152 is to classify the optimized voiceprint feature embedded in the global waveform feature vector of the guided acoustic wave signal through a classifier to obtain a classification result. In this step, a classifier is used to classify the optimized voiceprint feature embedded in the global waveform feature vector of the guided acoustic wave signal to determine whether the detected transformer has a fault. The classifier can be a machine learning model, such as a support vector machine (SVM) or a neural network, etc. It can learn the relationship between the features and labels of the sample data and can perform classification prediction on new inputs. The classification result indicating whether the transformer has a fault can help users detect and repair problems in a timely manner. Through feature distribution optimization and the application of the classifier, the voiceprint feature embedded in the global waveform feature vector of the guided acoustic wave signal can be transformed into a classification result for determining whether the transformer has a fault. Such a method can improve the accuracy and reliability of fault detection and provide strong support for the maintenance and management of transformers.

[0071] Here, each acoustic wave segment signal waveform feature vector in the sequence of the acoustic wave segment signal waveform feature vectors represents the image semantic features of the acoustic wave segment signal, while the acoustic texture feature parameter input vector represents the distribution information of the acoustic texture feature parameters along the sample direction. Thus, after passing the acoustic texture feature parameter input vector and the sequence of the acoustic wave segment signal waveform feature vectors through the feature embedding module, the temporal dynamic encoding of the image semantic features of the acoustic wave segment signals along the temporal direction can be performed based on the sample direction of the acoustic texture feature parameters. However, this also makes the spatio-temporal hybrid feature representation of the acoustic texture feature embedding-guided global waveform feature vector of the acoustic wave signal sparse in the cross-image semantic feature space distribution and temporal distribution, resulting in poor convergence of the probability density distribution of the regression probabilities of the respective eigenvalues of the acoustic texture feature embedding-guided global waveform feature vector of the acoustic wave signal when performing class probability regression mapping through the classifier, affecting the accuracy of the classification results obtained by the classifier. Therefore, preferably, the respective eigenvalues of the acoustic texture feature embedding-guided global waveform feature vector of the acoustic wave signal are optimized.

[0072] Correspondingly, in one example, optimizing the feature distribution of the acoustic texture feature embedding-guided global waveform feature vector of the acoustic wave signal to obtain an optimized acoustic texture feature embedding-guided global waveform feature vector of the acoustic wave signal includes: optimizing the feature distribution of the acoustic texture feature embedding-guided global waveform feature vector of the acoustic wave signal with the following optimization formula to obtain the optimized acoustic texture feature embedding-guided global waveform feature vector of the acoustic wave signal; where the optimization formula is:

[0073]

[0074] where V is the acoustic texture feature embedding-guided global waveform feature vector of the acoustic wave signal, v i and v j are the i-th and j-th eigenvalues of the acoustic texture feature embedding-guided global waveform feature vector of the acoustic wave signal, and is the global feature mean of the acoustic texture feature embedding-guided global waveform feature vector of the acoustic wave signal, exp{·} represents the exponential operation of a numerical value, and the exponential operation of the numerical value represents calculating the value of the natural exponential function with the numerical value as the power, and v′ i is the i-th eigenvalue of the optimized acoustic texture feature embedding-guided global waveform feature vector of the acoustic wave signal.

[0075] Specifically, aiming at the local probability density mismatch in the probability density distribution within the probability space caused by the sparse distribution of the acoustic feature-embedded guided acoustic signal global waveform feature vector in the high-dimensional feature space, through regularized global self-consistent class coding, to imitate the global self-consistent relationship of the coding behavior of the high-dimensional feature manifold of the acoustic feature-embedded guided acoustic signal global waveform feature vector in the probability space, so as to adjust the error landscape of the feature manifold in the high-dimensional open space domain, and achieve the self-consistent matching type class coding of the high-dimensional feature manifold of the acoustic feature-embedded guided acoustic signal global waveform feature vector for explicit probability space embedding, thereby enhancing the convergence of the probability density distribution of the regression probability of the acoustic feature-embedded guided acoustic signal global waveform feature vector, and improving the accuracy of the classification result obtained by the classifier.

[0076] Further, in step S152, the optimized acoustic feature-embedded guided acoustic signal global waveform feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the detected transformer has a fault, including: performing fully connected coding on the optimized acoustic feature-embedded guided acoustic signal global waveform feature vector using the fully connected layer of the classifier to obtain a coded classification feature vector; and inputting the coded classification feature vector into the Softmax classification function of the classifier to obtain the classification result.

[0077] That is, in the technical solution of the present invention, the labels of the classifier include that the detected transformer has a fault (the first label), and that the detected transformer has no fault (the second label), where the classifier determines which classification label the optimized acoustic feature-embedded guided acoustic signal global waveform feature vector belongs to through the softmax function. It should be noted that the first label p1 and the second label p2 here do not contain the concept set by humans. In fact, during the training process, the computer model does not have the concept of "whether the detected transformer has a fault". It only has two classification labels and the probabilities of the output features under these two classification labels, that is, the sum of p1 and p2 is one. Therefore, the classification result of whether the detected transformer has a fault is actually transformed into a binary class probability distribution that conforms to natural laws through the classification labels, and essentially uses the physical meaning of the natural probability distribution of the labels, rather than the linguistic text meaning of "whether the detected transformer has a fault".

[0078] It should be understood that the role of a classifier is to use given classes and known training data to learn classification rules and the classifier, and then classify (or predict) unknown data. Logistic regression, SVM, etc. are commonly used to solve binary classification problems. For multi-class classification problems, logistic regression or SVM can also be used, but it requires multiple binary classifications to form a multi-class classification, which is prone to errors and has low efficiency. Commonly used multi-class classification methods include the Softmax classification function.

[0079] It is worth mentioning that fully connected encoding is a process of mapping an input vector to a low-dimensional feature space. In the method of embedding acoustic feature guidance into the global waveform feature vector of acoustic signals, fully connected encoding is used to encode the optimized acoustic feature embedding into the global waveform feature vector of acoustic signals to generate an encoded classification feature vector. Fully connected encoding is usually implemented through a fully connected layer. A fully connected layer is a common layer type in neural networks, where each neuron is connected to all neurons in the previous layer. In a fully connected layer, each element in the input vector is multiplied by a weight and processed through an activation function to generate an encoded classification feature vector. The purpose of fully connected encoding is to map a high-dimensional input vector to a low-dimensional encoded feature vector by learning weight parameters, so as to retain key feature information and reduce redundancy. In this method, the fully connected encoding inputs the optimized acoustic feature embedding into the global waveform feature vector of acoustic signals into the fully connected layer. Through the learning of weight parameters and the processing of the activation function, an encoded classification feature vector is generated. This encoded classification feature vector will be used as the input of the subsequent classifier and classified through the Softmax classification function to obtain the final classification result, which is used to indicate whether there is a fault in the detected transformer. The purpose of fully connected encoding is to extract the most discriminative features in the input vector for better classification tasks.

[0080] In summary, the transformer defect detection method based on acoustic fingerprint detection according to the embodiments of the present invention is elucidated, which can achieve non-invasive transformer fault detection and identification.

[0081] Figure 6 FIG. is a block diagram of a transformer defect detection system 100 based on acoustic fingerprint detection according to an embodiment of the present invention. As Figure 6As shown, a transformer defect detection system 100 based on voiceprint detection according to an embodiment of the present invention includes: a sound wave signal acquisition module 110, configured to acquire a sound wave signal of a transformer to be detected collected by a sound wave sensor; a time-frequency analysis module 120, configured to perform time-frequency analysis on the sound wave signal to obtain voiceprint feature parameters; a local waveform feature extraction module 130, configured to extract local waveform features of the sound wave signal to obtain a sequence of sound wave segment signal waveform feature vectors; a voiceprint feature integration module 140, configured to integrate the voiceprint feature parameters into the sequence of sound wave segment signal waveform feature vectors to obtain a voiceprint feature embedded guiding sound wave signal global waveform feature vector; and a fault analysis module 150, configured to determine whether there is a fault in the transformer to be detected based on the voiceprint feature embedded guiding sound wave signal global waveform feature vector.

[0082] In one example, in the above-mentioned transformer defect detection system 100 based on voiceprint detection, the time-frequency analysis module 120 is configured to: perform time-frequency analysis on the sound wave signal based on short-time Fourier transform to obtain the voiceprint feature parameters.

[0083] Here, those skilled in the art can understand that the specific functions and operations of each module in the above-mentioned transformer defect detection system 100 based on voiceprint detection have been introduced in detail in the description of the Figures 1 to 5 transformer defect detection method based on voiceprint detection, and therefore, the repeated description thereof will be omitted.

[0084] As described above, the transformer defect detection system 100 based on voiceprint detection according to an embodiment of the present invention can be implemented in various wireless terminals, such as a server having a transformer defect detection algorithm based on voiceprint detection, etc. In one example, the transformer defect detection system 100 based on voiceprint detection according to an embodiment of the present invention can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the transformer defect detection system 100 based on voiceprint detection can be a software module in the operating system of the wireless terminal, or can be an application program developed for the wireless terminal; of course, the transformer defect detection system 100 based on voiceprint detection can also be one of the many hardware modules of the wireless terminal.

[0085] Alternatively, in another example, the transformer defect detection system 100 based on voiceprint detection and the wireless terminal can also be separate devices, and the transformer defect detection system 100 based on voiceprint detection can be connected to the wireless terminal through a wired and / or wireless network and transmit interaction information according to a predefined data format.

[0086] Figure 7This is an application scenario diagram of a transformer defect detection method based on voiceprint detection according to an embodiment of the present invention. As Figure 7 shown, in this application scenario, first, an acoustic wave signal of the transformer to be detected collected by an acoustic wave sensor is obtained (for example, Figure 7 D shown in Figure 7 ), and then the acoustic wave signal is input into a server (for example,

[0087] S shown in

[0088] ) deployed with a transformer defect detection algorithm based on voiceprint detection. Among them, the server can use the transformer defect detection algorithm based on voiceprint detection to process the acoustic wave signal to obtain a classification result for indicating whether the transformer to be detected has a fault.

[0089] Based on the same inventive concept, an embodiment of the present invention provides an electronic device, including:

[0090] One or more processors;

[0091] A storage unit for storing one or more programs, which, when executed by the one or more processors, can enable the one or more processors to implement the method described above. For details, reference can be made to the relevant records above, and no further elaboration will be provided here.

[0092] It is understandable that the above embodiments are merely exemplary embodiments adopted to illustrate the principles of the present invention. However, the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also regarded as the protection scope of the present invention.

Claims

1. A transformer defect detection method based on voiceprint detection, characterized in that: include: Acquiring an acoustic wave signal of the transformer under test collected by an acoustic wave sensor; Performing time-frequency analysis on the sound wave signal to obtain voiceprint feature parameters; Extracting local waveform features of the sound wave signal to obtain a sequence of waveform feature vectors of sound wave segment signals; Incorporating the voiceprint feature parameters into the sequence of waveform feature vectors of the sound wave segment signal to obtain a global waveform feature vector of the voiceprint feature embedding guided sound wave signal; and Based on the embedding of the voiceprint feature into the global waveform feature vector of the guided sound wave signal, it is determined whether the detected transformer has a fault.

2. The transformer defect detection method based on voiceprint detection according to claim 1 is characterized in that: Performing time-frequency analysis on the sound wave signal to obtain voiceprint feature parameters includes: performing time-frequency analysis based on short-time Fourier transform on the sound wave signal to obtain the voiceprint feature parameters.

3. The transformer defect detection method based on voiceprint detection according to claim 2 is characterized in that: Extracting local waveform features of the sound wave signal to obtain a sequence of waveform feature vectors of sound wave segment signals includes: Performing signal segmentation on the sound wave signal to obtain a sequence of sound wave segment signals; and The sequence of the sound wave segment signals is respectively passed through a sound wave waveform feature extractor based on a convolutional neural network model to obtain a sequence of the sound wave segment signal waveform feature vectors.

4. The transformer defect detection method based on voiceprint detection according to claim 3 is characterized in that: The acoustic waveform feature extractor based on the convolutional neural network model includes an input layer, a convolution layer, an activation layer, a pooling layer and an output layer.

5. The transformer defect detection method based on voiceprint detection according to claim 4 is characterized in that: The sequence of the sound wave segment signals is respectively passed through a sound wave waveform feature extractor based on a convolutional neural network model to obtain a sequence of waveform feature vectors of the sound wave segment signals, including: Each layer of the sound wave waveform feature extractor based on the convolutional neural network model is used to perform two-dimensional convolution processing, feature matrix-based mean pooling processing and nonlinear activation processing on the input data in the forward pass of the layer so that the last layer of the sound wave waveform feature extractor based on the convolutional neural network model outputs a sequence of waveform feature vectors of the sound wave segment signal, wherein the input of the first layer of the sound wave waveform feature extractor based on the convolutional neural network model is the sequence of the sound wave segment signals.

6. The transformer defect detection method based on voiceprint detection according to claim 5 is characterized in that: Integrating the voiceprint feature parameters into the sequence of waveform feature vectors of the sound wave segment signal to obtain a voiceprint feature embedding guided sound wave signal global waveform feature vector, including: Arranging the voiceprint feature parameters into a voiceprint feature parameter input vector; and The sequence of the sound wave segment signal waveform feature vectors and the voiceprint feature parameter input vector are input into a feature embedding module to obtain the voiceprint feature embedding-guided sound wave signal global waveform feature vector.

7. The transformer defect detection method based on voiceprint detection according to claim 6 is characterized in that: Inputting the sequence of the sound wave segment signal waveform feature vectors and the voiceprint feature parameter input vector into a feature embedding module to obtain the voiceprint feature embedding guided sound wave signal global waveform feature vector, including: The voiceprint feature parameter input vector passes through a feature extractor based on a fully convolutional network model to obtain a voiceprint feature time series feature vector; Linearly processing the voiceprint feature time series feature vector to obtain a linearly processed voiceprint feature time series feature vector; The sequence of the acoustic wave segment signal waveform feature vectors is subjected to linear processing to obtain a sequence of the acoustic wave segment signal waveform feature vectors after linear processing; Fusion of the linearly processed voiceprint feature time series feature vector and the linearly processed sound wave segment signal waveform feature vector sequence to obtain a voiceprint-sound wave linear initial fusion vector; Performing one-dimensional convolution processing on the sequence of waveform feature vectors of the sound wave segment signal to obtain a sequence of neighborhood correlation feature vectors of the sound wave time series; and The sequence of the acoustic wave time series neighborhood association feature vectors and the voiceprint-acoustic wave linear initial fusion vector are fused based on a splicing method to obtain the voiceprint feature embedding guidance acoustic wave signal global waveform feature vector.

8. The transformer defect detection method based on voiceprint detection according to claim 7 is characterized in that: Determining whether the detected transformer has a fault based on embedding the voiceprint feature into a global waveform feature vector of the guided sound wave signal includes: Performing feature distribution optimization on the global waveform feature vector of the voiceprint feature embedding guiding sound wave signal to obtain an optimized global waveform feature vector of the voiceprint feature embedding guiding sound wave signal; and The optimized voiceprint feature is embedded in the global waveform feature vector of the guided sound wave signal and passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the detected transformer has a fault.

9. A transformer defect detection system based on voiceprint detection, characterized in that: include: An acoustic wave signal acquisition module, used to acquire the acoustic wave signal of the transformer under test collected by the acoustic wave sensor; A time-frequency analysis module, used for performing time-frequency analysis on the sound wave signal to obtain voiceprint feature parameters; A local waveform feature extraction module, used for extracting local waveform features of the sound wave signal to obtain a sequence of waveform feature vectors of the sound wave segment signal; A voiceprint feature incorporation module, configured to incorporate the voiceprint feature parameters into the sequence of waveform feature vectors of the sound wave segment signal to obtain a global waveform feature vector of the sound wave signal guided by voiceprint feature embedding; as well as The fault analysis module is used to determine whether the detected transformer has a fault based on the global waveform feature vector of the guided sound wave signal embedded in the voiceprint feature.

10. The transformer defect detection system based on voiceprint detection according to claim 9, characterized in that: The time-frequency analysis module is used to perform time-frequency analysis on the sound wave signal based on short-time Fourier transform to obtain the voiceprint feature parameters.

Images