Method and device for extracting voiceprint features of transformer, storage medium and equipment

Through dual-spectrum analysis and weighted fusion processing technology, the stereo voiceprint characteristics of the transformer are extracted, which solves the problem of low accuracy in the transformer voiceprint recognition in the prior art, and realizes accurate analysis and health assessment of the transformer's operating status.

CN119091915APending Publication Date: 2024-12-06ZHANGJIAKOU POWER SUPPLY COMPANY OF STATE GRID JINBEI ELECTRIC POWER COMPANY +1
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410816617.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-24
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing transformer voiceprint recognition technology has low accuracy in the analysis of transformer operating status and is greatly affected by load current and noise interference.

Method used

The two-spectrum analysis algorithm is used to extend the sound signal to three-dimensional space, and the sound pattern spectrum is created through top-view projection and weighted fusion processing, and the voice pattern features, including texture and color features, and dimensionality reduction analysis is performed through the factor model.

Benefits of technology

It realizes accurate and comprehensive analysis of the operating status of the transformer, improves the identification accuracy, and provides valuable reference information for evaluating health status and equipment maintenance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091915A_ABST
    Figure CN119091915A_ABST
Patent Text Reader

Abstract

The invention discloses a method and device for extracting transformer voiceprint features, a storage medium and equipment, relates to the technical field of smart power grids, and mainly aims to solve the problem of low accuracy of transformer operation state analysis in the existing transformer voiceprint recognition technology. Analyzing and expanding each sound signal to a three-dimensional space by adopting a bispectrum analysis algorithm to obtain a three-dimensional spectrogram corresponding to each sound signal; overlooking projection processing is carried out on the three-dimensional spectrogram, and voiceprint contour maps corresponding to the sound signals are obtained; performing weighted fusion processing on the basis of the voiceprint contour map to obtain a stereo pattern spectrogram corresponding to the transformer; and performing feature extraction processing on the stereo pattern spectrogram to obtain voiceprint features corresponding to the transformer.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and in particular to a method and device for extracting transformer voiceprint features, a storage medium, and equipment. Background Art

[0002] With the continuous development of smart grids, transformers play a vital role in energy transmission and distribution systems. Transformer status monitoring and identification are essential to ensure the stable operation of the power grid. Traditional transformer status monitoring methods are often limited to monitoring parameters such as current and voltage. However, these monitoring methods sometimes cannot fully and accurately reflect the operating status of the transformer.

[0003] A transformer is an electrical device that transfers AC power from one circuit to another. Since the winding coils in the transformer are affected by electromotive force and the iron core is affected by magnetostrictive force, the transformer vibrates and makes sounds during operation. The vibration of the transformer contains rich information and can reflect the internal operating status of the transformer. Therefore, the operating status of the transformer can be analyzed by collecting the voiceprint data of the transformer. Generally, when performing transformer voiceprint recognition, the time-frequency analysis method is used to process the sound signal. However, the sound signal of the operating transformer will inevitably be affected by load current, noise interference, etc., which makes the accuracy of the existing transformer voiceprint recognition technology in analyzing the transformer operating status not high. Summary of the invention

[0004] In view of this, the present invention provides a method and device, storage medium and equipment for extracting transformer voiceprint features, the main purpose of which is to solve the problem that the existing transformer voiceprint recognition technology has low accuracy in analyzing the transformer operating status.

[0005] According to one aspect of the present invention, a method for extracting transformer voiceprint features is provided, comprising:

[0006] Acquire sound signals collected from multiple directions of the transformer, and use a bispectral analysis algorithm to analyze and expand each of the sound signals into a three-dimensional space to obtain a three-dimensional spectrogram corresponding to each of the sound signals;

[0007] Performing a top-down projection process on the three-dimensional sound spectrogram to obtain a voiceprint contour map corresponding to each of the sound signals;

[0008] Performing weighted fusion processing based on the voiceprint contour map to obtain a stereo voiceprint spectrogram corresponding to the transformer;

[0009] Feature extraction processing is performed on the stereo voiceprint spectrogram to obtain a voiceprint feature corresponding to the transformer.

[0010] Furthermore, the bispectral analysis algorithm is used to expand the analysis of each of the sound signals into a three-dimensional space to obtain a three-dimensional spectrogram corresponding to each of the sound signals, including:

[0011] Dividing the sound signal into a plurality of signal segments, and calculating the Fourier transform coefficients of each of the signal segments;

[0012] Performing triple correlation processing on each of the signal segments based on the Fourier transform coefficients to obtain a bispectral estimation of each of the signal segments;

[0013] The three-dimensional spectrogram corresponding to the sound signal is obtained by calculating an average value based on the bispectral estimation of each of the signal segments.

[0014] Furthermore, the weighted fusion processing is performed based on the voiceprint contour map to obtain a stereo voiceprint spectrogram corresponding to the transformer, including:

[0015] Segmenting each of the voiceprint contour maps respectively, and extracting voiceprint contour fragment maps corresponding to each of the voiceprint contour maps from the segmented results; each of the voiceprint contour fragment maps represents a non-overlapping area in the coordinate plane;

[0016] Obtaining weight values ​​of each of the voiceprint contour fragment graphs;

[0017] A weighted fusion process is performed based on the weight value and the voiceprint contour fragmentation map to obtain the stereo voiceprint spectrogram corresponding to the transformer.

[0018] Furthermore, the step of obtaining the weight value of each voiceprint contour fragment graph includes:

[0019] Calculating the time domain waveform energy value of the sound signal collected in each direction respectively;

[0020] Acquire a unilateral power spectrum of the sound signal, and calculate a unilateral power spectrum waveform energy value corresponding to the sound signal based on the unilateral power spectrum;

[0021] The weight values ​​corresponding to the respective voiceprint contour fragment graphs are determined based on the time domain waveform energy value and the single-sided power spectrum waveform energy value.

[0022] Furthermore, the performing feature extraction processing on the stereo voiceprint spectrogram to obtain the voiceprint feature corresponding to the transformer includes:

[0023] Extracting texture features of the stereo voiceprint spectrogram, and extracting color features of the stereo voiceprint spectrogram; and determining the voiceprint features corresponding to the transformer based on the extracted texture features and color features.

[0024] Further, the extracting of the texture features of the stereo soundprint spectrogram includes: calculating the corresponding roughness, contrast, directionality, linearity, regularity and coarseness based on the stereo soundprint spectrogram to obtain the texture features;

[0025] The extracting of the color features of the stereo voiceprint spectrogram includes: calculating the brightness information, color distribution range and color distribution symmetry on three channels of the color space based on the stereo voiceprint spectrogram to obtain the color features.

[0026] Furthermore, after performing feature extraction processing on the stereo voiceprint spectrogram to obtain the voiceprint feature corresponding to the transformer, the method further includes:

[0027] Extracting common factor variables from the voiceprint features through a factor model;

[0028] The common factor variables are determined as voiceprint features after dimensionality reduction, and the operating status of the transformer is analyzed using the voiceprint features after dimensionality reduction.

[0029] According to another aspect of the present invention, a device for extracting transformer voiceprint features is provided, comprising:

[0030] An acquisition and expansion module is used to acquire sound signals collected from multiple directions of the transformer, and use a bispectral analysis algorithm to analyze and expand each of the sound signals into a three-dimensional space to obtain a three-dimensional spectrogram corresponding to each of the sound signals;

[0031] A projection module, used for performing a top-down projection process on the three-dimensional sound spectrogram to obtain a voiceprint contour map corresponding to each of the sound signals;

[0032] A fusion processing module, used for performing weighted fusion processing based on the voiceprint contour map to obtain a stereo voiceprint spectrogram corresponding to the transformer;

[0033] The feature extraction module is used to perform feature extraction processing on the stereo voiceprint spectrogram to obtain the voiceprint features corresponding to the transformer.

[0034] Furthermore, the acquisition and expansion module is also used for:

[0035] Dividing the sound signal into a plurality of signal segments, and calculating the Fourier transform coefficients of each of the signal segments;

[0036] Performing triple correlation processing on each of the signal segments based on the Fourier transform coefficients to obtain a bispectral estimation of each of the signal segments;

[0037] The three-dimensional spectrogram corresponding to the sound signal is obtained by calculating an average value based on the bispectral estimation of each of the signal segments.

[0038] Furthermore, the fusion processing module includes a segmentation and extraction unit, a weight acquisition unit, and a fusion unit;

[0039] The segmentation and extraction unit is used to segment each of the voiceprint contour maps, and extract voiceprint contour fragment maps corresponding to each of the voiceprint contour maps from the segmentation results; each of the voiceprint contour fragment maps represents a non-overlapping area in the coordinate plane;

[0040] The weight acquisition unit is used to acquire the weight value of each of the voiceprint contour fragment graphs;

[0041] The fusion unit is used to perform weighted fusion processing based on the weight value and the voiceprint contour fragmentation map to obtain the stereo voiceprint spectrogram corresponding to the transformer.

[0042] Furthermore, the fusion processing module further includes a weight determination unit, which is used to:

[0043] Calculating the time domain waveform energy value of the sound signal collected in each direction respectively;

[0044] Acquire a unilateral power spectrum of the sound signal, and calculate a unilateral power spectrum waveform energy value corresponding to the sound signal based on the unilateral power spectrum;

[0045] The weight values ​​corresponding to the respective voiceprint contour fragment graphs are determined based on the time domain waveform energy value and the single-sided power spectrum waveform energy value.

[0046] Further, the feature extraction module includes a feature extraction unit and a feature determination unit;

[0047] The feature extraction unit is used to extract texture features of the stereo voiceprint spectrogram and extract color features of the stereo voiceprint spectrogram;

[0048] The feature determination unit is used to determine the voiceprint feature corresponding to the transformer based on the extracted texture feature and the color feature.

[0049] Furthermore, the feature extraction unit is also used for:

[0050] Based on the stereo sound pattern spectrogram, the corresponding roughness, contrast, directionality, linearity, regularity and coarseness are calculated respectively to obtain the texture feature;

[0051] Based on the stereo soundprint spectrogram, the brightness information, color distribution range and color distribution symmetry on the three channels of the color space are calculated respectively to obtain the color feature.

[0052] Furthermore, the device also includes a feature dimension reduction module, which is used to:

[0053] Extracting common factor variables from the voiceprint features through a factor model;

[0054] The common factor variables are determined as voiceprint features after dimensionality reduction, and the operating status of the transformer is analyzed using the voiceprint features after dimensionality reduction.

[0055] According to another aspect of the present invention, a storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to execute operations corresponding to the above-mentioned method for extracting transformer voiceprint features.

[0056] According to another aspect of the present invention, there is provided a device, comprising a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0057] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the method for extracting transformer voiceprint features as described above.

[0058] By means of the above technical solution, the technical solution provided by the embodiment of the present invention has at least the following advantages:

[0059] The present invention provides a method and device, storage medium, and equipment for extracting transformer voiceprint features. Compared with the prior art, the present invention obtains sound signals collected from multiple directions of the transformer, and uses a bispectral analysis algorithm to expand the analysis of each of the sound signals to three-dimensional space, thereby obtaining a three-dimensional sound spectrogram corresponding to each of the sound signals; performs a top-down projection process on the three-dimensional sound spectrogram to obtain a soundprint contour map corresponding to each of the sound signals; performs a weighted fusion process based on the soundprint contour map to obtain a stereo soundprint spectrogram corresponding to the transformer; performs feature extraction on the stereo soundprint spectrogram to obtain a soundprint feature corresponding to the transformer, so that the extracted soundprint feature can be used to analyze the operating status of the transformer. The present invention obtains all-round stereo soundprint information of the transformer based on the sound signal, and originally proposes a segmentation weighted reconstruction method to obtain a stereo soundprint spectrogram of the transformer, which can fuse the soundprint information from multiple angles, accurately and comprehensively reflect the status information of the transformer during operation, and provide valuable reference information for evaluating the health status of the transformer, reasonably arranging maintenance plans, and equipment maintenance.

[0060] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0062] Figure 1 A schematic flow chart of a method for extracting transformer voiceprint features provided by an embodiment of the present invention is shown;

[0063] Figure 2 A schematic diagram of a voiceprint contour map provided by an embodiment of the present invention is shown;

[0064] Figure 3 A schematic flow chart of another method for extracting transformer voiceprint features provided by an embodiment of the present invention is shown;

[0065] Figure 4 A schematic flow chart of another method for extracting transformer voiceprint features provided by an embodiment of the present invention is shown;

[0066] Figure 5 A schematic flow chart of another method for extracting transformer voiceprint features provided by an embodiment of the present invention is shown;

[0067] Figure 6 A schematic diagram of the structure of a device for extracting transformer voiceprint features provided by an embodiment of the present invention is shown;

[0068] Figure 7 A schematic structural diagram of a device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0069] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0070] The embodiment of the present invention provides a method for extracting transformer voiceprint features, such as Figure 1 As shown, the method includes:

[0071] 101. Acquire sound signals collected from multiple directions of the transformer, and use a bispectral analysis algorithm to analyze and expand each of the sound signals into a three-dimensional space to obtain a three-dimensional spectrogram corresponding to each of the sound signals;

[0072] In an embodiment of the present invention, the current execution end obtains sound signals collected from multiple directions of the transformer. Among them, the sound signal does not affect the normal inspection and maintenance operations of the transformer during collection, and the collection operation can be selected from the front (high-voltage bushing outlet side), back (low-voltage bushing outlet side), side (left or right) and bottom of the transformer. It should be noted that the sound sensors arranged in the above four directions cannot be in direct contact with the transformer box, and the distance from the surface of the transformer box should be kept at 20 to 50 cm. The sound signals collected in the four directions can more comprehensively reflect the operating status of the transformer, while minimizing the impact on on-site inspection and maintenance operations.

[0073] In an embodiment of the present invention, after the current execution end obtains the sound signal, the bispectrum analysis algorithm is used to expand the analysis of each sound signal to three-dimensional space, and a three-dimensional spectrogram corresponding to each sound signal is obtained. Among them, the bispectrum analysis algorithm has outstanding advantages such as good time-frequency performance, resolution and energy concentration. Traditional power spectrum analysis is only applicable to linear systems, while bispectrum analysis can detect and characterize the nonlinear characteristics of signals. The embodiment of the present invention uses bispectrum analysis to expand signal analysis to three-dimensional space, and the obtained three-dimensional spectrogram contains more transformer status information.

[0074] 102. Performing a top-down projection process on the three-dimensional sound spectrogram to obtain a voiceprint contour map corresponding to each of the sound signals;

[0075] In the embodiment of the present invention, the current execution end performs a top-down projection process on the three-dimensional sound spectrogram to obtain a voiceprint contour map corresponding to each of the sound signals. The voiceprint contour map reflects the change in the amplitude of the expanded sound signal in each dimension, such as Figure 2 shown.

[0076] 103. Perform weighted fusion processing based on the voiceprint contour map to obtain a stereo voiceprint spectrogram corresponding to the transformer;

[0077] In an embodiment of the present invention, the current execution end performs weighted fusion processing based on the voiceprint contour map to obtain a stereo voiceprint spectrogram corresponding to the transformer. The weighted fusion is to assign a weight to the signal data collected from each direction, and then perform weighted superposition on the multiple signal data collected from multiple directions based on the assigned weight to generate signal data containing more transformer operating status information. The weight assigned to the signal data collected from each direction reflects the contribution of the signal data to the generation of the stereo voiceprint spectrogram.

[0078] 104. Perform feature extraction processing on the stereo voiceprint spectrogram to obtain a voiceprint feature corresponding to the transformer.

[0079] In the embodiment of the present invention, the current execution end performs feature extraction processing on the stereo voiceprint spectrogram to obtain the voiceprint features corresponding to the transformer. Since the stereo voiceprint spectrogram reflects the state information of the transformer during operation in the voiceprint and color brightness of the stereo voiceprint spectrogram, when performing feature extraction, the color and texture features can be mainly extracted from the stereo voiceprint spectrogram, which is not specifically limited in the embodiment of the present invention.

[0080] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to make the collected sound signal more comprehensively reflect the operating status of the transformer, another method for extracting transformer voiceprint features is provided, such as Figure 3 As shown, the steps adopt a bispectral analysis algorithm to expand the analysis of each sound signal into a three-dimensional space to obtain a three-dimensional spectrogram corresponding to each sound signal, including:

[0081] 201. Divide the sound signal into a plurality of signal segments, and calculate the Fourier transform coefficient of each of the signal segments;

[0082] In the embodiment of the present invention, the current execution end divides the collected sound signal into a plurality of signal segments, for example, the collected sound signal a r (t) is divided into K segments, each of which contains M observation samples. Then, the Fourier transform coefficients of each signal segment are calculated to obtain A (k) r (λ), where λ=0,1,2,...,M / 2, k=0,1,2,...,K.

[0083] 202. Perform triple correlation processing on each of the signal segments based on the Fourier transform coefficients to obtain a bispectral estimation of each of the signal segments;

[0084] In the embodiment of the present invention, the current execution end performs triple correlation processing on each signal segment based on the Fourier transform coefficient calculated in step 201 to obtain a bispectral estimation of each signal segment. Specifically, the Fourier transform coefficient A (k)r The triple correlation of (λ)b (k) r (λ 1 ,λ 2 )for:

[0085]

[0086] Among them, Δ 0 =f s / N 0 , N 0 and L 1 Should satisfy M=(2L 1 +1)N 0 , f s is the signal sampling rate.

[0087] 203. Calculate an average value based on the bispectral estimation of each of the signal segments to obtain the three-dimensional spectrogram corresponding to the sound signal.

[0088] In the embodiment of the present invention, the current execution end calculates the average value based on the bispectral estimation of each signal segment, that is, the transformer sound signal a r (t) The corresponding three-dimensional spectrogram B r (ω 1 ,ω 2 ) is obtained by the average value of its K-segment bispectral estimation. The specific calculation formula is as follows:

[0089]

[0090] in,

[0091] It should be noted that, in the embodiment of the present invention, as in step 101, sound signals are collected from the front (high-voltage bushing outlet side), the back (low-voltage bushing outlet side), the side (either the left or the right), and the bottom of the transformer, which are respectively denoted as a r (t), a l (t), a m (t) and a n (t), it is necessary to use the method described in steps 201 to 203 to simultaneously calculate the three-dimensional spectrograms corresponding to the sound signals in the four directions, which are respectively denoted as B r (ω 1 ,ω 2 ), B l (ω 1 ,ω 2 ), B m (ω 1 ,ω 2 ) and B n (ω 1 ,ω2 ), which is not specifically limited in the embodiments of the present invention.

[0092] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to completely aggregate the voiceprint information of the same transformer at different positions into the same contour map according to the change of sound intensity, another method for extracting transformer voiceprint features is provided, such as Figure 4 As shown, the step of performing weighted fusion processing based on the voiceprint contour map to obtain a stereo voiceprint spectrogram corresponding to the transformer includes:

[0093] 301. Segment each of the voiceprint contour maps respectively, and extract voiceprint contour fragment maps corresponding to each of the voiceprint contour maps from the segmented results; each of the voiceprint contour fragment maps represents a non-overlapping area in a coordinate plane;

[0094] In the embodiment of the present invention, the current execution end performs segmentation processing on each voiceprint contour map according to the self-similarity of the voiceprint contour map, while ensuring that no information is lost. Then, the voiceprint contour fragment map corresponding to each voiceprint contour map is extracted from the segmented result, that is, the voiceprint contour fragment map at each position of the transformer is obtained. For example, for the three-dimensional spectrogram B recorded in step 203, r (ω 1 ,ω 2 ) The voiceprint contour map G is obtained by top-down projection processing r (ω 1 ,ω 2 ), then segmentation and feature extraction are performed to obtain the voiceprint contour fragment map S corresponding to the front position of the transformer r (ω 1 ,ω 2 ), the specific formula is as follows:

[0095]

[0096] The three-dimensional spectrogram B recorded in step 203 l (ω 1 ,ω 2 ) The voiceprint contour map G is obtained by top-down projection processing l (ω 1 ,ω 2 ), then segmentation and feature extraction are performed to obtain the voiceprint contour fragment map S corresponding to the position behind the transformer l (δ 1 ,δ 2 ), the specific formula is as follows:

[0097]

[0098] The three-dimensional spectrogram B recorded in step 203 m (ω 1 ,ω 2 ) The voiceprint contour map G is obtained by top-down projection processing m (ω 1 ,ω 2 ), then segmentation and feature extraction are performed to obtain the voiceprint contour fragment map S corresponding to the side orientation of the transformer m (ε 1 ,ε 2 ), the specific formula is as follows:

[0099]

[0100] The three-dimensional spectrogram B recorded in step 203 n (ω 1 ,ω 2 ) The voiceprint contour map G is obtained by top-down projection processing n (ω 1 ,ω 2 ), then segmentation and feature extraction are performed to obtain the voiceprint contour fragment map S corresponding to the position below the transformer n (φ 1 ,φ 2 ), the specific formula is as follows:

[0101]

[0102] It should be noted that the above S r (ω 1 ,ω 2 ), S l (δ 1 ,δ 2 ), S m (ε 1 ,ε 2 ) and S n (φ 1 ,φ 2 ) represent non-overlapping areas in the coordinate plane. r (ω 1 ,ω 2 ), S l (δ 1 ,δ 2 ), S m (ε 1 ,ε 2 ) and S n (φ 1 ,φ 2 ) can form a complete dual-spectrum voiceprint contour map.

[0103] 302. Obtaining weight values ​​of each of the voiceprint contour fragment graphs;

[0104] In the embodiment of the present invention, the current execution end obtains the weight value of each voiceprint height fragment map, such as obtaining the voiceprint height fragment map S in step 301. r (ω 1 ,ω 2 ), S l (δ 1 ,δ 2 ), S m (ε 1 ,ε 2 ) and S n (φ 1 ,φ 2 ) corresponds to the weight value α r , α l , α m and α n , the embodiments of the present invention are not specifically limited.

[0105] 303. Perform weighted fusion processing based on the weight value and the voiceprint contour fragmentation map to obtain the stereo voiceprint spectrogram corresponding to the transformer.

[0106] In the embodiment of the present invention, the current execution end performs weighted fusion processing based on the acquired weight value and the voiceprint contour fragment map to obtain the stereo voiceprint spectrogram corresponding to the transformer. The specific formula is as follows:

[0107] W(σ,τ)=α r S r (β 1 ,β 2 )+α l S l (δ 1 ,δ 2 )+α m S m (ε 1 ,ε 2 )+α n S n (φ 1 ,φ 2 )

[0108] Among them, W(σ, τ) is a stereo voiceprint spectrogram, which can be regarded as a main signal obtained after fusion. This signal can characterize the transformer state in all directions and from multiple angles, making it easy to deeply explore the information contained in the voiceprint from a holistic perspective.

[0109] Furthermore, as a refinement and extension of the specific implementation of the above embodiment, in order to further clarify the contribution of the stereo voiceprint spectrogram in each direction, another method for extracting transformer voiceprint features is provided, such as Figure 5As shown, the step of obtaining the weight value of each voiceprint contour fragment graph includes:

[0110] 401. Calculate the time domain waveform energy value of the sound signal collected in each direction respectively;

[0111] 402. Acquire a unilateral power spectrum of the sound signal, and calculate a unilateral power spectrum waveform energy value corresponding to the sound signal based on the unilateral power spectrum;

[0112] 403. Determine the weight value corresponding to each of the voiceprint contour fragments based on the time domain waveform energy value and the unilateral power spectrum waveform energy value.

[0113] In the embodiment of the present invention, the current execution end calculates the time domain waveform energy value of the sound signal collected in each direction, which is recorded as Next, obtain the unilateral power spectrum X of the sound signal respectively. g (f), then based on the unilateral power spectrum X g (f) Calculate the energy value of the unilateral power spectrum waveform corresponding to the sound signal, denoted as Finally, the current execution end determines the weight value α corresponding to each voiceprint isoheight fragment graph based on the time domain waveform energy value and the unilateral power spectrum waveform energy value g , the specific formula is as follows:

[0114]

[0115] Among them, t is the sampling value and f is the frequency value.

[0116] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to obtain more comprehensive features from the stereo voiceprint spectrogram, another method for extracting transformer voiceprint features is provided, the step of performing feature extraction processing on the stereo voiceprint spectrogram to obtain voiceprint features corresponding to the transformer, including:

[0117] Extracting texture features of the stereo voiceprint spectrogram, and extracting color features of the stereo voiceprint spectrogram;

[0118] In an embodiment of the present invention, the current execution end extracts the texture features of the stereo voiceprint spectrogram, including: calculating the corresponding roughness, contrast, directionality, linearity, regularity and coarseness based on the stereo voiceprint spectrogram to obtain the texture features;

[0119] In the embodiment of the present invention, the roughness indicates the granularity of the image texture pattern. The larger the granularity of the image texture pattern, the coarser the texture image is, and vice versa. Calculate the roughness T 1 The specific formula is as follows:

[0120]

[0121] Among them, p and q are the length and width of the image respectively, and V(i,j) is the optimal scale at the pixel point (i,j).

[0122] In the embodiment of the present invention, the contrast reflects the depth of the grooves of the image texture and the clarity of the image. The deeper the grooves, the greater the contrast and the clearer the image visual effect. Conversely, the shallower the grooves, the smaller the contrast and the more blurred the image. Calculate the contrast T 2 The specific formula is as follows:

[0123]

[0124] Among them, μ is the fourth-order moment and θ is the variance.

[0125] In the embodiment of the present invention, the directionality describes the concentration or dispersion of the texture along a certain direction, which is related to the shape and arrangement rules of the texture primitives. The more irregular the shape of the image, the larger the range of light and dark distribution, and the larger the directionality, and vice versa. Calculate the directionality T 3 The specific formula is as follows:

[0126]

[0127] in, e ζ is the number of histogram peaks, ζ is the histogram H D The peak value of (φ), ψ ζ is the peak quantization range, is ζ The quantized value in the maximum histogram.

[0128] In the embodiment of the present invention, linearity refers to the degree of deviation of the pixel spacing distance when calculating the local co-occurrence matrix. The larger the spacing between pixels in the image, the greater the linearity, and vice versa. Calculate the linearity T 4 The specific formula is as follows:

[0129]

[0130] Among them, P D Produces a θ×θ local directional co-occurrence matrix.

[0131] In the embodiment of the present invention, regularity measures the degree of regularity of the texture of an image. The more regular the image is, the closer the regularity value is to 1, otherwise the regularity value is closer to zero. Calculate the regularity T 5 The specific formula is as follows:

[0132] T 5 =1-R(ξ 1 +ξ 2 +ξ 3+ξ 4 )

[0133] Among them, R is the normalization factor, ξ 1 , 2 , 3 , 4 T 1 、T 2 、T 3 、T 4 The standard deviation of .

[0134] In the embodiment of the present invention, the coarseness varies with the shape, size and touch feeling of the image. The denser and more irregular the image is, the greater the coarseness is. Otherwise, the coarseness is smaller. 6 The specific formula is as follows:

[0135] T 6 =T 1 +T 2

[0136] In an embodiment of the present invention, the current execution end extracts the color features of the stereo voiceprint spectrogram, including: based on the stereo voiceprint spectrogram, respectively calculate the brightness information, color distribution range and color distribution symmetry on the three channels of the color space to obtain the color features. For an image, any color distribution therein can be represented by its moment. The first-order moment reflects the brightness information of the color block, the second-order moment reflects the distribution range of the color, and the third-order moment reflects the symmetry of the color distribution. The color moment can fully reflect its color distribution characteristics. In addition, for the three channels of the image color space HSV, the first-order moment, the second-order moment and the third-order moment are calculated respectively to obtain a 9-dimensional color feature. The specific formula is as follows:

[0137]

[0138]

[0139] Where i = h, s, v, respectively corresponding to the HSV channel, z i represents the first-order moment of channel i, ι i represents the second-order moment of channel i, η i represents the third-order moment of channel i, ρ ij represents the color value of the jth pixel on channel i, and Z is the number of pixels in the image.

[0140] In the embodiment of the present invention, the current execution end extracts the texture feature {T 1 、T 2 、T 3 、T 4 、T 5 、T 6} and color features {z i, i , η i , i=h,s,v} to determine the voiceprint feature corresponding to the transformer.

[0141] It should be noted that the voiceprint features extracted in the above feature extraction step may be correlated and redundant. In order to improve the pertinence and comprehensiveness of the voiceprint features, factor analysis can also be performed on the color and texture features to eliminate redundant data. Among them, factor analysis decomposes the original variables into linear combinations of factors by constructing a factor model, and uses a few related and independent potential factors to comprehensively reflect most of the information of all variables, so as to achieve the purpose of explaining complex research problems with a few variables.

[0142] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to eliminate redundant data in the voiceprint features and improve the pertinence and comprehensiveness of the features, another method for extracting transformer voiceprint features is provided, wherein after performing feature extraction processing on the stereo voiceprint spectrogram to obtain the voiceprint features corresponding to the transformer, the method further includes:

[0143] Extracting common factor variables from the voiceprint features through a factor model;

[0144] The common factor variables are determined as voiceprint features after dimensionality reduction, and the operating status of the transformer is analyzed using the voiceprint features after dimensionality reduction.

[0145] In the embodiment of the present invention, the specific formula for extracting common factor variables from voiceprint features through the factor model at the current execution end is as follows:

[0146] D κ×1 =Q κ×γ J γ×1 +I κ×1

[0147] Among them, D is the original variable, that is, the voiceprint feature that needs to be factor analyzed; Q is the factor loading matrix, and the elements in Q are called factor loadings; I is a special factor, which represents the part of the original variable that cannot be explained by the common factor; J is the common factor variable, that is, the voiceprint feature vector after dimensionality reduction; κ is the number of features that need to be factor analyzed, and γ is the number of features obtained after dimensionality reduction.

[0148] The current execution end uses factor analysis to reduce the dimension of the extracted color moment and texture features to obtain a stereo voiceprint fusion feature vector J reflecting the operating status of the transformer, and uses the reduced voiceprint feature vector J to analyze the operating status of the transformer.

[0149] The embodiment of the present invention provides a method for extracting transformer voiceprint features. Compared with the prior art, the present invention obtains sound signals collected from multiple directions of the transformer, and uses a bispectral analysis algorithm to expand the analysis of each of the sound signals to three-dimensional space, thereby obtaining a three-dimensional sound spectrogram corresponding to each of the sound signals; performs a top-down projection process on the three-dimensional sound spectrogram to obtain a soundprint contour map corresponding to each of the sound signals; performs a weighted fusion process based on the soundprint contour map to obtain a stereo soundprint spectrogram corresponding to the transformer; performs feature extraction on the stereo soundprint spectrogram to obtain a soundprint feature corresponding to the transformer, so that the extracted soundprint feature can be used to analyze the operating status of the transformer. The present invention obtains all-round stereo soundprint information of the transformer based on the sound signal, and originally proposes a segmentation weighted reconstruction method to obtain a stereo soundprint spectrogram of the transformer, which can fuse the soundprint information from multiple angles, accurately and comprehensively reflect the status information of the transformer during operation, and provide valuable reference information for evaluating the health status of the transformer, reasonably arranging maintenance plans and equipment maintenance.

[0150] As the above Figure 1 The embodiment of the present invention provides a device for extracting transformer voiceprint features, such as Figure 6 As shown, the device comprises:

[0151] The acquisition and expansion module 51 is used to acquire the sound signals collected from multiple directions of the transformer, and use the bispectral analysis algorithm to analyze and expand each of the sound signals into a three-dimensional space to obtain a three-dimensional spectrogram corresponding to each of the sound signals;

[0152] A projection module 52 is used to perform a top-down projection process on the three-dimensional sound spectrogram to obtain a voiceprint contour map corresponding to each of the sound signals;

[0153] A fusion processing module 53, configured to perform weighted fusion processing based on the voiceprint contour map to obtain a stereo voiceprint spectrogram corresponding to the transformer;

[0154] The feature extraction module 54 is used to perform feature extraction processing on the stereo voiceprint spectrogram to obtain the voiceprint feature corresponding to the transformer.

[0155] Furthermore, the acquisition and expansion module 51 is also used for:

[0156] Dividing the sound signal into a plurality of signal segments, and calculating the Fourier transform coefficients of each of the signal segments;

[0157] Performing triple correlation processing on each of the signal segments based on the Fourier transform coefficients to obtain a bispectral estimation of each of the signal segments;

[0158] The three-dimensional spectrogram corresponding to the sound signal is obtained by calculating an average value based on the bispectral estimation of each of the signal segments.

[0159] Furthermore, the fusion processing module 53 includes a segmentation and extraction unit, a weight acquisition unit, and a fusion unit;

[0160] The segmentation and extraction unit is used to segment each of the voiceprint contour maps, and extract voiceprint contour fragment maps corresponding to each of the voiceprint contour maps from the segmentation results; each of the voiceprint contour fragment maps represents a non-overlapping area in the coordinate plane;

[0161] The weight acquisition unit is used to acquire the weight value of each of the voiceprint contour fragment graphs;

[0162] The fusion unit is used to perform weighted fusion processing based on the weight value and the voiceprint contour fragmentation map to obtain the stereo voiceprint spectrogram corresponding to the transformer.

[0163] Furthermore, the fusion processing module 53 also includes a weight determination unit, which is used to:

[0164] Calculating the time domain waveform energy value of the sound signal collected in each direction respectively;

[0165] Acquire a unilateral power spectrum of the sound signal, and calculate a unilateral power spectrum waveform energy value corresponding to the sound signal based on the unilateral power spectrum;

[0166] The weight values ​​corresponding to the respective voiceprint contour fragment graphs are determined based on the time domain waveform energy value and the single-sided power spectrum waveform energy value.

[0167] Further, the feature extraction module 54 includes a feature extraction unit and a feature determination unit;

[0168] The feature extraction unit is used to extract texture features of the stereo voiceprint spectrogram and extract color features of the stereo voiceprint spectrogram;

[0169] The feature determination unit is used to determine the voiceprint feature corresponding to the transformer based on the extracted texture feature and the color feature.

[0170] Furthermore, the feature extraction unit is also used for:

[0171] Based on the stereo sound pattern spectrogram, the corresponding roughness, contrast, directionality, linearity, regularity and coarseness are calculated respectively to obtain the texture feature;

[0172] Based on the stereo soundprint spectrogram, the brightness information, color distribution range and color distribution symmetry on the three channels of the color space are calculated respectively to obtain the color feature.

[0173] Furthermore, the device also includes a feature dimension reduction module, which is used to:

[0174] Extracting common factor variables from the voiceprint features through a factor model;

[0175] The common factor variables are determined as voiceprint features after dimensionality reduction, and the operating status of the transformer is analyzed using the voiceprint features after dimensionality reduction.

[0176] The embodiment of the present invention provides a device for extracting transformer voiceprint features. Compared with the prior art, the present invention obtains sound signals collected from multiple directions of the transformer, and uses a bispectral analysis algorithm to expand the analysis of each of the sound signals to three-dimensional space, thereby obtaining a three-dimensional sound spectrogram corresponding to each of the sound signals; performs a top-down projection process on the three-dimensional sound spectrogram to obtain a soundprint contour map corresponding to each of the sound signals; performs a weighted fusion process based on the soundprint contour map to obtain a stereo soundprint spectrogram corresponding to the transformer; performs feature extraction on the stereo soundprint spectrogram to obtain a soundprint feature corresponding to the transformer, so that the extracted soundprint feature can be used to analyze the operating status of the transformer. The present invention obtains all-round stereo soundprint information of the transformer based on the sound signal, and originally proposes a segmentation weighted reconstruction method to obtain a stereo soundprint spectrogram of the transformer, which can fuse the soundprint information from multiple angles, accurately and comprehensively reflect the status information of the transformer during operation, and provide valuable reference information for evaluating the health status of the transformer, reasonably arranging maintenance plans and equipment maintenance.

[0177] According to an embodiment of the present invention, a storage medium is provided, wherein the storage medium stores at least one executable instruction, and the computer executable instruction can execute the method for extracting transformer voiceprint features in any of the above method embodiments.

[0178] Figure 7 A schematic diagram of the structure of a device provided according to an embodiment of the present invention is shown, and the specific embodiment of the present invention does not limit the specific implementation of the device.

[0179] like Figure 7 As shown, the device may include: a processor (processor) 602 , a communication interface (Communications Interface) 604 , a memory (memory) 606 , and a communication bus 608 .

[0180] The processor 602 , the communication interface 604 , and the memory 606 communicate with each other via a communication bus 608 .

[0181] The communication interface 604 is used to communicate with other devices such as clients or other servers.

[0182] The processor 602 is used to execute the program 610, and specifically can execute the relevant steps of the above-mentioned method for extracting transformer voiceprint features.

[0183] Specifically, the program 610 may include program codes, which include computer operation instructions.

[0184] The processor 602 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0185] The memory 606 is used to store the program 610. The memory 606 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0186] The program 610 may be specifically configured to enable the processor 602 to perform the following operations:

[0187] Acquire sound signals collected from multiple directions of the transformer, and use a bispectral analysis algorithm to analyze and expand each of the sound signals into a three-dimensional space to obtain a three-dimensional spectrogram corresponding to each of the sound signals;

[0188] Performing a top-down projection process on the three-dimensional sound spectrogram to obtain a voiceprint contour map corresponding to each of the sound signals;

[0189] Performing weighted fusion processing based on the voiceprint contour map to obtain a stereo voiceprint spectrogram corresponding to the transformer;

[0190] Feature extraction processing is performed on the stereo voiceprint spectrogram to obtain a voiceprint feature corresponding to the transformer.

[0191] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0192] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for extracting transformer voiceprint features, characterized in that: include: Acquire sound signals collected from multiple directions of the transformer, and use a bispectral analysis algorithm to analyze and expand each of the sound signals into a three-dimensional space to obtain a three-dimensional spectrogram corresponding to each of the sound signals; Performing a top-down projection process on the three-dimensional sound spectrogram to obtain a voiceprint contour map corresponding to each of the sound signals; Performing weighted fusion processing based on the voiceprint contour map to obtain a stereo voiceprint spectrogram corresponding to the transformer; Feature extraction processing is performed on the stereo voiceprint spectrogram to obtain a voiceprint feature corresponding to the transformer.

2. The method according to claim 1, characterized in that The bispectral analysis algorithm is used to expand the analysis of each sound signal into a three-dimensional space to obtain a three-dimensional spectrogram corresponding to each sound signal, including: Dividing the sound signal into a plurality of signal segments, and calculating the Fourier transform coefficients of each of the signal segments; Performing triple correlation processing on each of the signal segments based on the Fourier transform coefficients to obtain a bispectral estimation of each of the signal segments; The three-dimensional spectrogram corresponding to the sound signal is obtained by calculating an average value based on the bispectral estimation of each of the signal segments.

3. The method according to claim 1, characterized in that The weighted fusion processing is performed based on the voiceprint contour map to obtain a stereo voiceprint spectrogram corresponding to the transformer, comprising: Segmenting each of the voiceprint contour maps respectively, and extracting voiceprint contour fragment maps corresponding to each of the voiceprint contour maps from the segmented results; each of the voiceprint contour fragment maps represents a non-overlapping area in the coordinate plane; Obtaining weight values ​​of each of the voiceprint contour fragment graphs; A weighted fusion process is performed based on the weight value and the voiceprint contour fragmentation map to obtain the stereo voiceprint spectrogram corresponding to the transformer.

4. The method according to claim 3, characterized in that The obtaining of the weight values ​​of each of the voiceprint contour fragment graphs includes: Calculating the time domain waveform energy value of the sound signal collected in each direction respectively; Acquire a unilateral power spectrum of the sound signal, and calculate a unilateral power spectrum waveform energy value corresponding to the sound signal based on the unilateral power spectrum; The weight values ​​corresponding to the respective voiceprint contour fragment graphs are determined based on the time domain waveform energy value and the single-sided power spectrum waveform energy value.

5. The method according to claim 1, characterized in that The performing feature extraction processing on the stereo voiceprint spectrogram to obtain a voiceprint feature corresponding to the transformer includes: Extracting texture features of the stereo voiceprint spectrogram, and extracting color features of the stereo voiceprint spectrogram; The voiceprint feature corresponding to the transformer is determined based on the extracted texture feature and the color feature.

6. The method according to claim 5, characterized in that The extracting of the texture features of the stereo soundprint spectrogram includes: calculating the corresponding roughness, contrast, directionality, linearity, regularity and coarseness based on the stereo soundprint spectrogram to obtain the texture features; The extracting of the color features of the stereo voiceprint spectrogram includes: calculating the brightness information, color distribution range and color distribution symmetry on three channels of the color space based on the stereo voiceprint spectrogram to obtain the color features.

7. The method according to any one of claims 1 to 6, characterized in that: After performing feature extraction processing on the stereo voiceprint spectrogram to obtain the voiceprint feature corresponding to the transformer, the method further includes: Extracting common factor variables from the voiceprint features through a factor model; The common factor variables are determined as voiceprint features after dimensionality reduction, and the operating status of the transformer is analyzed using the voiceprint features after dimensionality reduction.

8. A device for extracting transformer voiceprint features, characterized in that: include: An acquisition and expansion module is used to acquire sound signals collected from multiple directions of the transformer, and use a bispectral analysis algorithm to analyze and expand each of the sound signals into a three-dimensional space to obtain a three-dimensional spectrogram corresponding to each of the sound signals; A projection module, used for performing a top-down projection process on the three-dimensional sound spectrogram to obtain a voiceprint contour map corresponding to each of the sound signals; A fusion processing module, used for performing weighted fusion processing based on the voiceprint contour map to obtain a stereo voiceprint spectrogram corresponding to the transformer; The feature extraction module is used to perform feature extraction processing on the stereo voiceprint spectrogram to obtain the voiceprint features corresponding to the transformer.

9. A storage medium, wherein at least one executable instruction is stored in the storage medium, and the executable instruction executes an operation corresponding to the method for extracting transformer voiceprint features according to any one of claims 1 to 7.

10. A device comprising a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the method for extracting transformer voiceprint features according to any one of claims 1 to 7.

Citation Information

Cited By

  • Motor fault diagnosis method based on voiceprint feature extraction and transfer learning

    CN119667476A