Electrical equipment electric leakage identification method and identification device based on voiceprint signal

Abstract

The invention provides an electrical equipment electric leakage identification method and identification device based on voiceprint signals. The method comprises the following steps: processing an acquired sound signal by using a variational mode decomposition algorithm to obtain a plurality of intrinsic mode functions, selecting three key intrinsic mode functions from the plurality of intrinsic mode functions to respectively serve as gray values of three color channels of red, green and blue of an image, and generating a voiceprint image of the sound signal; extracting texture features of the voiceprint image through a gray level co-occurrence matrix, and performing depth feature extraction and classification identification on the texture features by adopting a CNN-SVM model to obtain an electric leakage detection result; and when the electric leakage detection result shows that an electric leakage fault occurs, the position of a fault source is determined by calculating the time delay difference of the acoustic signals transmitted to the plurality of acoustic signal sensors. According to the scheme, the problem that in the prior art, accurate electric leakage fault source positioning cannot be provided, so that maintenance and fault removal time is prolonged is solved.

Description

Technical Field

[0001] The present application relates to the technical field of power system automation, and in particular to a method, an identification device, a computer-readable storage medium and an electronic device for identifying leakage of electrical equipment based on voiceprint signals. Background Art

[0002] Leakage detection of electrical equipment in distribution networks is crucial to the safe operation of power systems. With the widespread application of AC / DC hybrid power supply networks, the diversification of electrical equipment types and the increase in system complexity, traditional leakage detection methods have gradually exposed their limitations. Traditional detection methods mainly rely on current difference, leakage current detection or voltage imbalance analysis. These methods are easily affected by noise interference, nonlinear loads and other factors when dealing with complex electrical environments, resulting in low detection accuracy of leakage faults, especially in small leakage and early fault detection, which are prone to missed detection or false detection. In addition, with the continuous expansion of the scale of distribution systems and the wide distribution of electrical equipment, the rapid location of leakage faults also faces huge challenges. Traditional methods often cannot provide accurate fault source location, resulting in extended maintenance and troubleshooting time, affecting the stability and safety of power grid operation. Therefore, there is an urgent need for a detection method that can quickly locate the fault source to ensure the efficient operation of the distribution network and the safety and stability of electrical equipment. Summary of the invention

[0003] The main purpose of the present application is to provide a method, an identification device, a computer-readable storage medium and an electronic device for identifying leakage of electrical equipment based on voiceprint signals, so as to at least solve the problem that the prior art cannot provide accurate leakage fault source positioning, resulting in prolonged maintenance and troubleshooting time.

[0004] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for identifying leakage of electrical equipment based on voiceprint signals is provided, wherein the electrical equipment is an AC / DC hybrid power supply and distribution network electrical equipment, comprising: using a variational mode decomposition algorithm to process the collected acoustic signal to obtain multiple intrinsic mode functions, and selecting three key intrinsic mode functions from the multiple intrinsic mode functions; using the three key intrinsic mode functions as the grayscale values ​​of the red, green and blue color channels of the image respectively to generate a voiceprint image of the acoustic signal; extracting the texture features of the voiceprint image through a grayscale co-occurrence matrix, wherein the texture features include contrast features, energy features and entropy features; using a CNN-SVM model to perform deep feature extraction and classification recognition on the texture features to obtain a leakage detection result; when the leakage detection result indicates that a leakage fault has occurred, calculating the time delay difference of the acoustic signal propagating to multiple acoustic signal sensors, and using the time delay difference to determine the location of the fault source.

[0005] Optionally, the collected acoustic signal is processed using a variational mode decomposition algorithm to obtain a plurality of intrinsic mode functions, and three key intrinsic mode functions are selected from the plurality of intrinsic mode functions, including: applying formula 1 The original acoustic signal is decomposed into a set of modal components, where K is the number of modal components, {u k} is a set of modal components, {ω k} is the center frequency corresponding to the modal component, u k (t) is the kth modal component, ω k is the center frequency of the kth modal component, represents the derivative operation of the signal, j represents the imaginary unit, t is the time variable, * represents the convolution operation, and δ(t) is the Dirac function; for the formula 1, the Lagrangian multiplier λ and the penalty term α are introduced to construct the Lagrangian function Iterative optimization algorithm is used to solve u k ,ω k and λ(t) until the change of the modal component is less than a preset threshold, wherein λ(t) is the Lagrange multiplier function and f(t) is the original acoustic signal; analyzing the characteristics of all the modal components, and selecting the three modal components containing the key information of the signal as the three key intrinsic mode functions.

[0006] Optionally, the three key intrinsic mode functions are used as grayscale values ​​of the red, green and blue color channels of the image respectively to generate the voiceprint image of the acoustic signal, including: normalizing each of the key intrinsic mode functions; creating an RGB image framework, the RGB image is composed of red, green and blue color channels, each of the color channels represents information of different colors of the RGB image; using formula P n The voiceprint image is constructed by using the formula (i, j)=B((n-1)M×N / 2+M(i-1)+j), wherein P(.) represents the pixel grayscale value at the corresponding position of the RGB image, B(.) represents the normalized amplitude of the key intrinsic mode function, the size of the RGB image is M×N, i=1:M, j=1:N, and the normalized amplitudes of the three key intrinsic mode functions are respectively used as the grayscale values ​​of the red (R), green (G), and blue (B) color channels of the RGB image.

[0007] Optionally, the grayscale co-occurrence matrix construction process includes: selecting a reference point (x, y) and a co-occurrence point (x+a, y+b) on the voiceprint image, so that the grayscale values ​​of the reference point and the co-occurrence point constitute a grayscale value combination (g1, g2); let the grayscale value level of the voiceprint image be k, transform the coordinates of the reference point and the co-occurrence point to obtain various grayscale value combinations (g1, g2), and the combination of the grayscale value combinations (g1, g2) has k 2 traverse all the reference points in the voiceprint image, count the occurrence frequency of each gray value combination (g1, g2), and obtain frequency statistics results; organize the frequency statistics results into a square matrix, normalize the square matrix and convert it into a probability matrix P (g1, g2), the probability matrix is ​​the initial gray level co-occurrence matrix, wherein the rows and columns of the initial gray level co-occurrence matrix correspond to the gray values ​​g1 and g2 in the voiceprint image respectively; use the formula The initial gray level co-occurrence matrix is ​​normalized, wherein p'(g1, g2, a, b) is the gray level co-occurrence matrix after normalization, a and b are the distance difference values ​​of the gray values ​​g1 and g2 in the voiceprint image, N' is the number of gray levels, and R is the normalization factor.

[0008] Optionally, extracting the texture features of the voiceprint image through a gray level co-occurrence matrix includes: using the formula The contrast feature quantity is extracted from the gray level co-occurrence matrix using the formula The energy feature quantity is extracted from the gray level co-occurrence matrix using the formula The entropy feature is extracted from the gray-level co-occurrence matrix, wherein i and j traverse the rows and columns of the gray-level co-occurrence matrix, respectively representing a gray-level value in the gray-level co-occurrence matrix, and p(i,j) is an element in the gray-level co-occurrence matrix, representing the probability of a pixel pair with gray-level values ​​​​i and j appearing at a given distance difference value (a, b).

[0009] Optionally, the construction process of the CNN-SVM model includes: constructing a CNN model, the CNN model includes an input layer, a convolution layer, an activation layer, a pooling layer and a fully connected layer, wherein the input layer is used to take the voiceprint image as input, the convolution layer is used to automatically learn the texture features of the voiceprint image through a convolution kernel, the activation layer is used to increase the nonlinear expression ability of the network, the pooling layer is used to reduce the size of the voiceprint image, extract high-level features of the voiceprint image, and reduce computational complexity and prevent overfitting, the fully connected layer uses a ReLU activation function to convert the two-dimensional feature map output by the convolution layer into a one-dimensional vector to obtain a feature vector; the feature vector is input into the SVM model for training to obtain the CNN-SVM model.

[0010] Optionally, using the delay difference to determine the location of the fault source includes: using formula d i =v×(t i -t0) calculate the distance d between the fault source and the i-th sensor i , where t0 is the time when the leakage fault occurs, t i is the time when the i-th sensor receives the acoustic signal, v is the speed of sound; solve the objective function Calculate the coordinates of the fault source in space, and use the coordinates of the fault source in space as the position of the fault source, where P is the position of the fault source, P i is the position of the i-th sensor, and D is the number of the acoustic signal sensors.

[0011] In order to achieve the above object, according to another aspect of the present application, a device for identifying leakage of electrical equipment based on voiceprint signals is provided, comprising:

[0012] A processing unit, configured to process the collected acoustic signal using a variational mode decomposition algorithm to obtain a plurality of intrinsic mode functions, and select three key intrinsic mode functions from the plurality of intrinsic mode functions;

[0013] A generating unit, configured to use the three key intrinsic mode functions as grayscale values ​​of the red, green and blue color channels of the image, respectively, to generate a voiceprint image of the acoustic signal;

[0014] An extraction unit, configured to extract texture features of the voiceprint image through a gray level co-occurrence matrix, wherein the texture features include a contrast feature quantity, an energy feature quantity, and an entropy feature quantity;

[0015] A classification and recognition unit, used for performing deep feature extraction and classification and recognition on the texture features using a CNN-SVM model to obtain a leakage detection result;

[0016] The calculation unit is used to calculate the time delay difference of the acoustic signal propagating to multiple acoustic signal sensors when the leakage detection result indicates that a leakage fault occurs, and determine the location of the fault source using the time delay difference.

[0017] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute any one of the methods for identifying leakage of electrical equipment based on voiceprint signals.

[0018] According to another aspect of the present application, an electronic device is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include means for executing any one of the electrical equipment leakage identification methods based on voiceprint signals.

[0019] The technical solution of the present application is applied. First, the collected acoustic signal is processed using the variational mode decomposition algorithm to obtain multiple intrinsic mode functions, and three key intrinsic mode functions are selected from the multiple intrinsic mode functions; the three key intrinsic mode functions are used as the grayscale values ​​of the red, green and blue color channels of the image to generate the voiceprint image of the acoustic signal; then, the texture features of the voiceprint image are extracted through the grayscale co-occurrence matrix, and the texture features include contrast features, energy features and entropy features; the CNN-SVM model is used to perform deep feature extraction and classification recognition on the texture features to obtain the leakage detection result; finally, when the leakage detection result indicates that a leakage fault has occurred, the delay difference of the acoustic signal propagating to multiple acoustic signal sensors is calculated, and the delay difference is used to determine the location of the fault source. This solution solves the problem that the prior art cannot provide accurate leakage fault source positioning, resulting in extended maintenance and troubleshooting time. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings constituting part of the present application are used to provide a further understanding of the present application. The exemplary embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0021] Figure 1 A hardware structure block diagram of a mobile terminal for executing a method for identifying leakage of electrical equipment based on a voiceprint signal provided in an embodiment of the present application is shown;

[0022] Figure 2 A schematic diagram of a flow chart of a method for identifying leakage of electrical equipment based on voiceprint signals according to an embodiment of the present application is shown;

[0023] Figure 3 A CNN-SVM recognition algorithm structure diagram of an electrical equipment leakage recognition method based on a voiceprint signal provided in an embodiment of the present application is shown;

[0024] Figure 4 A structural block diagram of an electrical equipment leakage identification device based on a voiceprint signal provided according to an embodiment of the present application is shown.

[0025] The reference numerals are as follows:

[0026] 102, processor; 104, memory; 106, transmission device; 108, input and output devices. DETAILED DESCRIPTION

[0027] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0028] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] As introduced in the background technology, when dealing with complex electrical environments, the existing technology is easily affected by factors such as noise interference and nonlinear loads, resulting in low detection accuracy of leakage faults, especially in the detection of small leakages and early faults, which is prone to missed detections or false detections. In order to solve the problem of being unable to provide accurate fault source positioning, resulting in prolonged maintenance and troubleshooting time, the embodiments of the present application provide an electrical equipment leakage identification method, identification device, computer-readable storage medium and electronic device based on voiceprint signals.

[0031] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0032] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 1 is a hardware structure block diagram of a mobile terminal of an electrical equipment leakage identification method based on voiceprint signals according to an embodiment of the present invention. Figure 1As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0033] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the display method of device information in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The transmission device 106 is used to receive or send data via a network. The above-mentioned specific examples of the network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0034] In this embodiment, a method for identifying leakage of electrical equipment based on voiceprint signals that runs on a mobile terminal, a computer terminal or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0035] Figure 2FIG. 1 is a flow chart diagram of a method for identifying leakage of electrical equipment based on voiceprint signals according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0036] Step S201 : Processing the collected acoustic signal using a variational mode decomposition algorithm to obtain a plurality of intrinsic mode functions, and selecting three key intrinsic mode functions from the plurality of the intrinsic mode functions.

[0037] Specifically, variational mode decomposition (VMD) is a signal processing technique used to decompose complex signals into a set of intrinsic mode functions (IMFs), which can better reflect the intrinsic characteristics of the signal. In the embodiment of the present application, the variational mode decomposition algorithm is used to pre-process the collected acoustic signal, the purpose of which is to separate and extract the acoustic features related to the leakage fault.

[0038] Among the multiple intrinsic mode functions decomposed by the variational mode decomposition algorithm, not all intrinsic mode functions are directly related to leakage faults. Some intrinsic mode functions may mainly reflect environmental noise, acoustic modes of normal equipment operation, or other interference signals. Therefore, it is very important to select the three key intrinsic mode functions that are most related to leakage faults from all the decomposed intrinsic mode functions.

[0039] Step S202: Using the three key intrinsic mode functions as grayscale values ​​of the red, green and blue color channels of the image, respectively, to generate a voiceprint image of the acoustic signal.

[0040] Specifically, the amplitudes of these three key intrinsic mode functions need to be converted into grayscale values ​​in the image. Grayscale values ​​are usually between 0 and 255, with 0 representing the darkest (black) and 255 representing the brightest (white). The mapping process standardizes the amplitude range of the key intrinsic mode functions to the interval [0, 255], so that the signal intensity information in each modal component is converted into the brightness information of the color channel in the image.

[0041] When generating a voiceprint image, the first key intrinsic mode function is mapped to the red (R) channel of the image, the second is mapped to the green (G) channel, and the third is mapped to the blue (B) channel. The grayscale value of each channel represents the intensity information of the corresponding key intrinsic mode function. Through this color channel allocation, the spectrum and intensity characteristics of the signal can be intuitively represented in the image. Once the amplitude of the key intrinsic mode function is converted to the grayscale value of the RGB channel, the next step is to construct the voiceprint image. The voiceprint image is a texture image that can visually represent the characteristics of the acoustic signal, including spectral distribution, intensity, and time characteristics. In the RGB space, different color combinations can form a rich visual effect, thereby highlighting specific patterns or features in the acoustic signal, which facilitates subsequent feature extraction and pattern recognition.

[0042] Step S203, extracting texture features of the voiceprint image through a gray level co-occurrence matrix, wherein the texture features include contrast feature quantity, energy feature quantity and entropy feature quantity.

[0043] Specifically, the gray-level co-occurrence matrix is ​​a method to quantify image texture, which counts the frequency of gray-level value combinations between all pixels in the image and their neighboring pixels. For a given image and a specific offset vector (i.e., the relative position between pixels), each element in the gray-level co-occurrence matrix represents the number of times the gray-level values ​​of two pixels appear at the same time. Since the voiceprint image encodes the characteristics of the acoustic signal in the gray-level value, the gray-level co-occurrence matrix can reveal the distribution pattern of these acoustic characteristics in the image. By calculating texture features such as contrast, energy, and entropy, the key information in the voiceprint image can be further extracted, providing richer feature inputs for the subsequent CNN-SVM model to identify and classify different acoustic modes, including normal states and leakage fault states.

[0044] Step S204: Use the CNN-SVM model to perform deep feature extraction and classification recognition on the above texture features to obtain leakage detection results.

[0045] Specifically, a convolutional neural network (CNN) is a neural network specifically designed to process input data (such as images) with a grid structure. In an embodiment of the present application, CNN is used to extract deeper features from voiceprint images. CNN can automatically learn and extract local and global features of images through a multi-layer structure such as convolutional layers, activation layers, and pooling layers. For voiceprint images, CNN can identify and extract texture patterns related to leakage faults, such as edges, shapes, and distributions.

[0046] Support Vector Machine (SVM) is a supervised learning model used for classification and regression analysis. In the embodiment of the present application, SVM is used to classify the feature vector extracted by CNN to determine whether the voiceprint image contains the characteristics of leakage fault. SVM distinguishes samples of different categories by finding an optimal hyperplane that maximizes the interval between different categories, thereby achieving accurate classification of new samples.

[0047] The CNN-SVM model combines the automatic feature extraction capability of deep learning with the classification and recognition capability of SVM. CNN can automatically learn highly abstract features from raw image data, while SVM can make decisions based on these features to determine whether there is a leakage fault. This combination can improve the generalization and robustness of the model, enabling the model to accurately detect leakage faults in complex and changing electrical environments.

[0048] Step S205 , when the leakage detection result indicates that a leakage fault occurs, the time delay difference of the acoustic signal propagating to the plurality of acoustic signal sensors is calculated, and the location of the fault source is determined using the time delay difference.

[0049] Specifically, when the fault source generates acoustic signals, these signals will propagate in all directions through the medium of the distribution network (such as air or insulating materials). Since different sensors are at different distances from the fault source, the time it takes for the acoustic signal to reach each sensor will also be different. This time difference is called delay. By recording the exact time of arrival of the acoustic signal by the sensor, the delay difference between each pair of sensors can be calculated. Once the delay differences between different sensors are calculated, these delay differences can be used to determine the location of the fault source.

[0050] In the above embodiment, first, the collected acoustic signal is processed by using the variational mode decomposition algorithm to obtain multiple intrinsic mode functions, and three key intrinsic mode functions are selected from the multiple intrinsic mode functions; the three key intrinsic mode functions are used as the grayscale values ​​of the red, green and blue color channels of the image respectively to generate the voiceprint image of the acoustic signal; then, the texture features of the voiceprint image are extracted by the grayscale co-occurrence matrix, and the texture features include contrast feature quantity, energy feature quantity and entropy feature quantity; the CNN-SVM model is used to perform deep feature extraction and classification recognition on the texture features to obtain the leakage detection result; finally, when the leakage detection result indicates that a leakage fault occurs, the delay difference of the acoustic signal propagating to multiple acoustic signal sensors is calculated, and the location of the fault source is determined by using the delay difference. This solves the problem that the prior art cannot provide accurate leakage fault source positioning, resulting in extended maintenance and troubleshooting time.

[0051] In the specific implementation process, the variational mode decomposition algorithm is used to process the collected acoustic signal to obtain multiple intrinsic mode functions, and three key intrinsic mode functions are selected from the multiple intrinsic mode functions, including: Applying formula 1 The original acoustic signal is decomposed into a set of modal components, where K is the number of modal components, {u k} is a set of the above modal components, {ω k} is the center frequency corresponding to the above modal components, u k (t) is the kth modal component, ω k is the center frequency of the kth modal component mentioned above, represents the derivative operation of the signal, j represents the imaginary unit, * represents the convolution operation, and δ(t) is the Dirac function; for the above formula 1, the Lagrangian multiplier λ and the penalty term α are introduced to construct the Lagrangian function Iterative optimization algorithm is used to solve u k ,ω k and λ(t) until the change of the above modal components is less than a preset threshold, wherein λ(t) is the Lagrange multiplier function and f(t) is the original acoustic signal; analyzing the characteristics of all the above modal components, and selecting the three modal components containing the key information of the signal as the three key intrinsic mode functions.

[0052] Specifically, the basic idea of ​​the variational mode decomposition algorithm is to decompose the signal into a set of modal components by constructing a variational model, and the central frequency ω of each modal component is k Adaptively estimate through optimization process. The core idea of ​​the algorithm is to modulate the signal so that the spectrum of each mode is concentrated near zero frequency. Then, a band-limited filter is used to extract the frequency components of each mode. By minimizing the bandwidth of each modal component, the modal components and their corresponding center frequencies ω are solved. k .

[0053] In formula one In , the goal of the variational modal decomposition algorithm is expressed, which is to minimize the spectral bandwidth of each modal component while satisfying the constraints and ensuring that the sum of all modal components is equal to the original sound signal. In order to solve the above constrained optimization problem, the variational modal decomposition algorithm introduces Lagrangian multipliers and penalty parameters to construct the Lagrangian function. The Lagrangian function combines the objective function and constraints, and solves the modal components, center frequencies, and Lagrangian multipliers through iterative optimization algorithms, such as the Alternating Direction Method of Multipliers (ADMM), until the change in the modal component is less than the preset threshold. This process ensures the stability and convergence of the decomposition.

[0054] After obtaining all the modal components, analyze the characteristics of each modal component and select three modal components containing key information of the signal as key intrinsic mode functions. This is usually based on the center frequency, energy distribution, time characteristics, etc. of the modal components to determine which modal components can best reflect the abnormal conditions of the signal, such as leakage faults. The selected key intrinsic mode functions will be used for subsequent voiceprint image construction and feature extraction to achieve more accurate fault detection and location.

[0055] In the specific implementation process, the three key intrinsic mode functions are used as the grayscale values ​​of the red, green and blue color channels of the image respectively to generate the voiceprint image of the acoustic signal, including: normalizing each of the key intrinsic mode functions; creating an RGB image framework, the RGB image is composed of red, green and blue color channels, each of which represents the information of different colors of the RGB image; using formula P n The voiceprint image is constructed by using the formula (i,j)=B((n-1)M×N / 2+M(i-1)+j), wherein P(.) represents the grayscale value of the pixel at the corresponding position of the RGB image, and B(.) represents the normalized amplitude of the key intrinsic mode function. The size of the RGB image is M×N, i=1:M, j=1:N, and the normalized amplitudes of the three key intrinsic mode functions are respectively used as the grayscale values ​​of the red (R), green (G), and blue (B) color channels of the RGB image.

[0056] Specifically, first, the three key intrinsic mode functions extracted from the acoustic signal need to be normalized. Normalization is to map the amplitudes of these key intrinsic mode functions to the range of image grayscale values, usually between 0 and 255. The purpose of this is to ensure that the information of each modal component can be displayed in the image with appropriate brightness, avoiding certain signal features from being ignored or misinterpreted due to large amplitude differences.

[0057] Creating an M×N RGB image frame means that the image will have M rows and N columns of pixels, and each pixel contains information about the three color channels of red, green, and blue. Each channel of the RGB image uses 8 bits (256 grayscale levels) to represent color information, so the grayscale value range of each channel is 0 to 255.

[0058] Using P nThe voiceprint image is constructed by the formula (i,j)=B((n-1)M×N / 2+M(i-1)+j), which maps the normalized key intrinsic mode function amplitude to the pixel grayscale value of the RGB image. This formula means that for each pixel in the image, its RGB color value will be determined by the normalized amplitude of the three key intrinsic mode functions at the corresponding position. In this way, the information of different frequencies and intensities in the acoustic signal is encoded in the color of the image, forming a voiceprint image.

[0059] In a specific implementation process, the construction process of the gray-level co-occurrence matrix includes: selecting a reference point (x, y) and a co-occurrence point (x+a, y+b) on the voiceprint image, so that the grayscale values ​​of the reference point and the co-occurrence point form a grayscale value combination (g1, g2); let the grayscale value level of the voiceprint image be k, transform the coordinates of the reference point and the co-occurrence point to obtain various grayscale value combinations (g1, g2), and the combination of the grayscale value combination (g1, g2) has k 2 traverse all the reference points in the voiceprint image, count the occurrence frequency of each gray value combination (g1, g2), and obtain the frequency statistics result; organize the frequency statistics result into a square matrix, normalize the square matrix and convert it into a probability matrix P (g1, g2), the probability matrix is ​​the initial gray level co-occurrence matrix, wherein the rows and columns of the initial gray level co-occurrence matrix correspond to the gray values ​​g1 and g2 in the voiceprint image respectively; adopt the formula The above initial gray level co-occurrence matrix is ​​normalized, where p'(g1, g2, a, b) is the gray level co-occurrence matrix after normalization, a and b are the distance difference values ​​of the gray level values ​​g1 and g2 in the voiceprint image, N' is the number of gray levels, and R is the normalization factor.

[0060] Specifically, a grayscale co-occurrence matrix is ​​constructed to quantify the distribution and spatial relationship of gray levels in an image, thereby extracting statistics that describe the texture features of the image. On the voiceprint image, first select a reference point (x, y), and then select a co-occurrence point (x+a, y+b) that is a certain distance away from the reference point in a specific direction. Here, a and b define the spatial relationship between the two points, which can be horizontal, vertical, or diagonal offsets. The grayscale values ​​of the reference point and the co-occurrence point constitute a grayscale value combination (g1, g2), where g1 is the grayscale value of the reference point, and g2 is the grayscale value of the co-occurrence point. Since the grayscale value of the voiceprint image has k levels, there are a total of k possible grayscale value combinations. 2 Traverse the entire voiceprint image, for each pair of reference points and co-occurrence points, record the number of times their gray value combination (g1, g2) appears. This process needs to be performed for every possible reference point in the image, and finally a result of counting the frequency of occurrence of all gray value combinations is obtained.

[0061] The above frequency statistics are organized into a square matrix, where the rows of the matrix correspond to the grayscale value g1, the columns correspond to the grayscale value g2, and each element in the matrix represents the frequency of occurrence of the combination (g1, g2). Then, this square matrix is ​​normalized, that is, divided by the total number of pixels in the image to convert it into a probability matrix P(g1, g2). The normalized matrix elements represent the probability of the grayscale value combination (g1, g2) appearing in the image. Use the formula The probability matrix P(g1,g2) is further normalized to ensure that the sum of the matrix elements is equal to 1. Normalization is done by dividing each matrix element by a sum or factor (R) to ensure that the sum of the matrix elements statistically represents a probability distribution.

[0062] After completing the above steps, the normalized gray-level co-occurrence matrix p'(g1, g2, a, b) is a statistical matrix that quantitatively describes the relative frequency and spatial relationship of gray values ​​in the image. This matrix can be used to calculate a variety of texture features, such as contrast, energy, and entropy, to further reveal the intrinsic structure and characteristics of the voiceprint image, and provide valuable feature input for electrical equipment leakage detection based on voiceprint signals. In practical applications, since the construction of the gray-level co-occurrence matrix depends on the selection of reference points and co-occurrence points, different selection strategies and distance differential values ​​a and b will lead to different matrices. Therefore, multiple gray-level co-occurrence matrices are usually constructed in different directions and offsets, and then the average or maximum value features are extracted from these matrices to fully reflect the texture characteristics of the image.

[0063] In the specific implementation process, the texture features of the above voiceprint image are extracted through the gray level co-occurrence matrix, including: using the formula The above contrast feature quantity is extracted from the above gray level co-occurrence matrix using the formula The above energy feature quantity is extracted from the above gray-level co-occurrence matrix using the formula The above entropy feature is extracted from the above gray-level co-occurrence matrix, where i and j traverse the rows and columns of the above gray-level co-occurrence matrix, respectively representing a gray-level value in the above gray-level co-occurrence matrix, and p(i,j) is an element in the above gray-level co-occurrence matrix, indicating the probability of a pixel pair with gray-level values ​​​​i and j appearing at a given distance difference value (a, b).

[0064] Specifically, the texture features of the voiceprint image are extracted through the gray-level co-occurrence matrix, including contrast features, energy features and entropy features, which are used to quantify the texture properties of the image and provide feature input for subsequent image recognition and classification. The contrast feature, also known as contrast, directly reflects the contrast of the brightness of the image pixel values. It is proportional to the depth and clarity of the texture grooves. The larger its value, the deeper the grooves and the clearer the effect. The energy feature is an indicator of the stability of the grayscale change of the texture image, reflecting the fineness of the texture and the grayscale distribution. The energy value is equal to the square sum of the element values ​​in the matrix. The larger the energy value, the more stable the regular change of the texture. The entropy feature is an indicator of the amount of image information, reflecting the complexity of the texture in the image, and its value is proportional to the complexity of the image. When the pixel values ​​in the gray-level co-occurrence matrix show great randomness or all their values ​​are the same, the entropy value is the largest and the image is more complex.

[0065] By calculating the contrast feature, energy feature and entropy feature of the grayscale co-occurrence matrix, the texture structure of the voiceprint image can be quantified, providing key feature input for the image-based leakage detection method. These feature quantities can not only reveal the distribution and contrast of grayscale values ​​in the image, but also reflect the complexity and information content of the image, which is important for distinguishing normal images from faulty images. In the embodiment of the present application, these feature quantities are used to train and optimize the CNN-SVM model to improve the accuracy and robustness of leakage detection.

[0066] In the specific implementation process, the construction process of the above-mentioned CNN-SVM model includes: constructing a CNN model, the above-mentioned CNN model includes an input layer, a convolution layer, an activation layer, a pooling layer and a fully connected layer, wherein the above-mentioned input layer is used to take the above-mentioned voiceprint image as input, the above-mentioned convolution layer is used to automatically learn the above-mentioned texture features of the above-mentioned voiceprint image through a convolution kernel, the above-mentioned activation layer is used to increase the nonlinear expression ability of the network, the above-mentioned pooling layer is used to reduce the size of the above-mentioned voiceprint image, extract the high-level features of the above-mentioned voiceprint image, and at the same time reduce the computational complexity and prevent overfitting, and the above-mentioned fully connected layer uses a ReLU activation function to convert the two-dimensional feature map output by the above-mentioned convolution layer into a one-dimensional vector to obtain a feature vector; the above-mentioned feature vector is input into the SVM model for training to obtain the above-mentioned CNN-SVM model.

[0067] Specifically, the process of building a CNN-SVM model is to combine two machine learning models, Convolutional Neural Network (CNN) and Support Vector Machine (SVM), to efficiently extract features from voiceprint images and classify them. CNN is a feedforward neural network that contains convolution calculations and has a deep structure. The CNN network not only supports multi-layer neural networks, but also can learn feature parameters and classifiers at the same time. The structure of the CNN network is similar to a multi-layer perceptron, with a feedforward structure and the characteristics of weight sharing and local perception in image analysis. Its basic network structure includes 5 parts: input layer, convolution layer, pooling layer, fully connected layer and output layer.

[0068] The convolution layer is the core layer of the convolutional neural network. It processes the input image data through convolution operations to extract local features of the image. The essence of the convolution kernel is a weight matrix. In addition to the weight matrix itself, the number and step size of the convolution kernel are also important parameters. The corresponding number of convolution kernels are selected according to the different dimensions of the extracted feature matrix. The input size is proportionally reduced by the step size. That is, the convolution layer can change the size of the image and the number of channels of the image, as shown in the formula As shown, w1 is the width of the input image, h1 is the height, c1 and m represent the number of channels, k is the convolution kernel size, s is the operation step size, and p is the zero padding size, then w2, h2, and c2 are the width, height, and number of channels of the output image.

[0069] The pooling layer downsamples the image input by the previous layer, reduces the size of the matrix, and thus reduces the number of nodes in the fully connected layer, thereby reducing network parameters and avoiding overfitting. Commonly used pooling methods include average pooling and maximum pooling. The pooling layer only changes the size of the image, does not affect the depth of the three-dimensional matrix, and maintains the invariance of local features. Its mathematical model is as follows: As shown, w1 is the width of the input image, h1 is the height, c1 represents the number of channels, k is the convolution kernel size, s is the operation step size, p is the zero padding size, then w2, h2, c2 are the width, height and number of channels of the output image.

[0070] The fully connected layer is a traditional feedforward neural network structure in which each neuron is fully connected to all neurons in the previous layer. Its main function is to summarize the information input from the previous layer and act as a "classifier" in the neural network to classify the features extracted from the feature map of the previous layer. The fully connected layer usually uses the ReLU activation function and is usually located in the last few layers of the network to convert the two-dimensional feature map output by the convolutional layer into a one-dimensional vector to achieve an end-to-end learning process.

[0071] The output layer is the last layer of the neural network. Its function is to pass the output of the previous layer to the output function, usually using the logistic regression function (i.e. Softmax regression). The function of the Softmax function is to convert the output of the fully connected layer into a value between 0 and 1 and output the result in the form of probability. It is often used in classification tasks in convolutional neural networks. The Softmax function is defined as follows: As shown, where pi is the probability of the category output, the numerator is the exponential transformation of the output corresponding to the category, and the denominator is the sum of the exponential transformations of the output of the corresponding category, and n represents the category type.

[0072] VGG16 is a relatively mature convolutional neural network in the current CNN network. The VGG16 architecture is used as the CNN part of the network model, as shown in the formula input→[(conv×2→pool)×2]→[(conv×3→pool)×3]→FC×3. VGG16 uses the softmax function as the final output of the VGG16 architecture, but since the softmax classifier only performs probability conversion on the input of the previous layer, it does not have the ability to learn independently and cannot effectively improve the generalization ability of the image classification recognition model. Therefore, SVM is used instead of the softmax function to recalculate the values ​​in the final output layer.

[0073] The output feature vector of the CNN model is input into the SVM model for training. SVM is a supervised learning model used for classification and regression tasks. Its core is to find a hyperplane that can maximize the class interval to distinguish samples of different categories. In the application embodiment, SVM is used to learn the decision boundary between leakage and normal state based on the feature vector extracted by CNN. The training process of the SVM model involves finding the best classification hyperplane so that the model can accurately distinguish between voiceprint images with leakage and without leakage. Through training, the SVM model can learn the key features that describe leakage faults, which will be used for subsequent leakage detection and fault source location. After the SVM model training is completed, it can predict whether there is leakage based on the feature vector of the new voiceprint image.

[0074] In summary, the construction process of the CNN-SVM model is to automatically learn and extract the texture features of the voiceprint image through CNN, and then map these feature vectors to the decision space through SVM to achieve the classification of the leakage status of electrical equipment. This method that combines deep learning and traditional machine learning can not only automatically extract key features in the image, but also use the strong classification ability of SVM to perform accurate classification tasks, thereby improving the accuracy and efficiency of leakage detection.

[0075] In the specific implementation process, the above delay difference is used to determine the location of the fault source, including: using formula d i =v×(ti -t0) calculate the distance d between the above fault source and the i-th sensor i , where t0 is the time when the above leakage fault occurs, t i is the time when the i-th sensor receives the above sound signal, v is the speed of sound; solve the objective function Calculate the coordinates of the fault source in space, and use the coordinates of the fault source in space as the position of the fault source, where P is the position of the fault source, P i is the position of the i-th sensor, and D is the number of the acoustic signal sensors.

[0076] Specifically, the location of the sound source is determined based on the time difference of the sound signal arriving at different sensors, and the distance is estimated using the speed of sound and the time difference. Then, the time difference data of multiple sensors is used to solve the location of the sound source using triangulation or overdetermined equations. First, the relative distance between the fault source and each sensor is estimated by detecting the time difference of the sound signal arriving at each sensor. This is the basis for positioning using the time difference. Next, the objective function is used To determine the exact coordinates of the fault source in space, the objective function aims to minimize the difference between the actual distance and the estimated distance. By solving this optimization problem, an accurate position estimate of the fault source can be obtained. Generally, gradient descent, least squares method or more complex nonlinear optimization algorithms can be used to solve such problems.

[0077] The structure diagram of CNN-SVM recognition algorithm is as follows Figure 3 As shown in the figure, for a voiceprint image, the key feature information is first extracted through the convolution layer, and then the feature information is "purified" using the pooling layer. This process is repeated many times, and finally a multi-dimensional vector describing the internal features of the image is obtained. Since these multi-dimensional feature vectors cannot be directly used as input for SVM, data conversion is required. For this purpose, a fully connected neural network is added to the model to convert the final pooling result into a one-dimensional feature vector, which is then input into the SVM for model training.

[0078] In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the method for identifying leakage of electrical equipment based on voiceprint signals of the present application will be described in detail below in combination with specific embodiments.

[0079] The embodiment of the present application also provides an electrical equipment leakage identification device based on voiceprint signals. It should be noted that the electrical equipment leakage identification device based on voiceprint signals in the embodiment of the present application can be used to execute the electrical equipment leakage identification method based on voiceprint signals provided in the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation in hardware, or a combination of software and hardware, is also possible and conceivable.

[0080] The following is an introduction to the electrical equipment leakage identification device based on voiceprint signals provided in an embodiment of the present application.

[0081] Figure 4 It is a structural block diagram of an electrical equipment leakage identification device based on voiceprint signals according to an embodiment of the present application.

[0082] like Figure 4 As shown, the device comprises:

[0083] The processing unit 10 is used to process the collected acoustic signal using a variational mode decomposition algorithm to obtain a plurality of intrinsic mode functions, and select three key intrinsic mode functions from the plurality of the intrinsic mode functions.

[0084] Specifically, variational mode decomposition (VMD) is a signal processing technique used to decompose complex signals into a set of intrinsic mode functions (IMFs), which can better reflect the intrinsic characteristics of the signal. In the embodiment of the present application, the variational mode decomposition algorithm is used to pre-process the collected acoustic signal, the purpose of which is to separate and extract the acoustic features related to the leakage fault.

[0085] Among the multiple intrinsic mode functions decomposed by the variational mode decomposition algorithm, not all intrinsic mode functions are directly related to leakage faults. Some intrinsic mode functions may mainly reflect environmental noise, acoustic modes of normal equipment operation, or other interference signals. Therefore, it is very important to select the three key intrinsic mode functions that are most related to leakage faults from all the decomposed intrinsic mode functions.

[0086] The generating unit 20 is used to use the three key intrinsic mode functions as grayscale values ​​of the red, green and blue color channels of the image respectively to generate the voiceprint image of the acoustic signal.

[0087] Specifically, the amplitudes of these three key intrinsic mode functions need to be converted into grayscale values ​​in the image. Grayscale values ​​are usually between 0 and 255, with 0 representing the darkest (black) and 255 representing the brightest (white). The mapping process standardizes the amplitude range of the key intrinsic mode functions to the interval [0, 255], so that the signal intensity information in each modal component is converted into the brightness information of the color channel in the image.

[0088] When generating a voiceprint image, the first key intrinsic mode function is mapped to the red (R) channel of the image, the second is mapped to the green (G) channel, and the third is mapped to the blue (B) channel. The grayscale value of each channel represents the intensity information of the corresponding key intrinsic mode function. Through this color channel allocation, the spectrum and intensity characteristics of the signal can be intuitively represented in the image. Once the amplitude of the key intrinsic mode function is converted to the grayscale value of the RGB channel, the next step is to construct the voiceprint image. The voiceprint image is a texture image that can visually represent the characteristics of the acoustic signal, including spectral distribution, intensity, and time characteristics. In the RGB space, different color combinations can form a rich visual effect, thereby highlighting specific patterns or features in the acoustic signal, which facilitates subsequent feature extraction and pattern recognition.

[0089] The extraction unit 30 is used to extract the texture features of the voiceprint image through the gray level co-occurrence matrix, and the texture features include contrast feature quantity, energy feature quantity and entropy feature quantity.

[0090] Specifically, the gray-level co-occurrence matrix is ​​a method to quantify image texture, which counts the frequency of gray-level value combinations between all pixels in the image and their neighboring pixels. For a given image and a specific offset vector (i.e., the relative position between pixels), each element in the gray-level co-occurrence matrix represents the number of times the gray-level values ​​of two pixels appear at the same time. Since the voiceprint image encodes the characteristics of the acoustic signal in the gray-level value, the gray-level co-occurrence matrix can reveal the distribution pattern of these acoustic characteristics in the image. By calculating texture features such as contrast, energy, and entropy, the key information in the voiceprint image can be further extracted, providing richer feature inputs for the subsequent CNN-SVM model to identify and classify different acoustic modes, including normal states and leakage fault states.

[0091] The classification and recognition unit 40 is used to perform deep feature extraction and classification and recognition on the above texture features using a CNN-SVM model to obtain leakage detection results.

[0092] Specifically, a convolutional neural network (CNN) is a neural network specifically designed to process input data (such as images) with a grid structure. In an embodiment of the present application, CNN is used to extract deeper features from voiceprint images. CNN can automatically learn and extract local and global features of images through a multi-layer structure such as convolutional layers, activation layers, and pooling layers. For voiceprint images, CNN can identify and extract texture patterns related to leakage faults, such as edges, shapes, and distributions.

[0093] Support Vector Machine (SVM) is a supervised learning model used for classification and regression analysis. In the embodiment of the present application, SVM is used to classify the feature vector extracted by CNN to determine whether the voiceprint image contains the characteristics of leakage fault. SVM distinguishes samples of different categories by finding an optimal hyperplane that maximizes the interval between different categories, thereby achieving accurate classification of new samples.

[0094] The CNN-SVM model combines the automatic feature extraction capability of deep learning with the classification and recognition capability of SVM. CNN can automatically learn highly abstract features from raw image data, while SVM can make decisions based on these features to determine whether there is a leakage fault. This combination can improve the generalization and robustness of the model, enabling the model to accurately detect leakage faults in complex and changing electrical environments.

[0095] The calculation unit 50 is used to calculate the time delay difference of the acoustic signal propagating to multiple acoustic signal sensors when the leakage detection result indicates that a leakage fault occurs, and determine the location of the fault source using the time delay difference.

[0096] Specifically, when the fault source generates acoustic signals, these signals will propagate in all directions through the medium of the distribution network (such as air or insulating materials). Since different sensors are at different distances from the fault source, the time it takes for the acoustic signal to reach each sensor will also be different. This time difference is called delay. By recording the exact time of arrival of the acoustic signal by the sensor, the delay difference between each pair of sensors can be calculated. Once the delay differences between different sensors are calculated, these delay differences can be used to determine the location of the fault source.

[0097] The processing unit includes:

[0098] Decomposition module, used to apply formula 1 The original acoustic signal is decomposed into a set of modal components, where K is the number of modal components, {u k} is a set of the above modal components, {ω k} is the center frequency corresponding to the above modal components, u k (t) is the kth modal component, ω k is the center frequency of the kth modal component mentioned above, represents the derivative operation of the signal, j represents the imaginary unit, * represents the convolution operation, and δ(t) is the Dirac function;

[0099] The solution module is used to introduce the Lagrangian multiplier λ and the penalty term α for the above formula 1 to construct the Lagrangian function Iterative optimization algorithm is used to solve u k ,ω k and λ(t), until the change of the above modal component is less than a preset threshold, wherein λ(t) is the Lagrange multiplier function and f(t) is the original acoustic signal;

[0100] The selection module is used to analyze the characteristics of all the above modal components and select the three above modal components containing key information of the signal as the three above key eigenmode functions.

[0101] Specifically, the basic idea of ​​the variational mode decomposition algorithm is to decompose the signal into a set of modal components by constructing a variational model, and the central frequency ω of each modal component is k Adaptively estimate through optimization process. The core idea of ​​the algorithm is to modulate the signal so that the spectrum of each mode is concentrated near zero frequency. Then, a band-limited filter is used to extract the frequency components of each mode. By minimizing the bandwidth of each modal component, the modal components and their corresponding center frequencies ω are solved. k .

[0102] In formula one In , the goal of the variational modal decomposition algorithm is expressed, which is to minimize the spectral bandwidth of each modal component while satisfying the constraints and ensuring that the sum of all modal components is equal to the original sound signal. In order to solve the above constrained optimization problem, the variational modal decomposition algorithm introduces Lagrangian multipliers and penalty parameters to construct the Lagrangian function. The Lagrangian function combines the objective function and constraints, and solves the modal components, center frequencies, and Lagrangian multipliers through iterative optimization algorithms, such as the Alternating Direction Method of Multipliers (ADMM), until the change in the modal component is less than the preset threshold. This process ensures the stability and convergence of the decomposition.

[0103] After obtaining all the modal components, analyze the characteristics of each modal component and select three modal components containing key information of the signal as key intrinsic mode functions. This is usually based on the center frequency, energy distribution, time characteristics, etc. of the modal components to determine which modal components can best reflect the abnormal conditions of the signal, such as leakage faults. The selected key intrinsic mode functions will be used for subsequent voiceprint image construction and feature extraction to achieve more accurate fault detection and location.

[0104] The above generation unit includes:

[0105] A processing module, used for normalizing each of the above key intrinsic mode functions;

[0106] A creation module is used to create a framework of an RGB image, wherein the RGB image is composed of three color channels: red, green, and blue, and each of the color channels represents information of different colors of the RGB image;

[0107] Building blocks for using formula P n The voiceprint image is constructed by using the formula (i,j)=B((n-1)M×N / 2+M(i-1)+j), wherein P(.) represents the grayscale value of the pixel at the corresponding position of the RGB image, and B(.) represents the normalized amplitude of the key intrinsic mode function. The size of the RGB image is M×N, i=1:M, j=1:N, and the normalized amplitudes of the three key intrinsic mode functions are respectively used as the grayscale values ​​of the red (R), green (G), and blue (B) color channels of the RGB image.

[0108] Specifically, first, the three key intrinsic mode functions extracted from the acoustic signal need to be normalized. Normalization is to map the amplitudes of these key intrinsic mode functions to the range of image grayscale values, usually between 0 and 255. The purpose of this is to ensure that the information of each modal component can be displayed in the image with appropriate brightness, avoiding certain signal features from being ignored or misinterpreted due to large amplitude differences.

[0109] Creating an M×N RGB image frame means that the image will have M rows and N columns of pixels, and each pixel contains information about the three color channels of red, green, and blue. Each channel of the RGB image uses 8 bits (256 grayscale levels) to represent color information, so the grayscale value range of each channel is 0 to 255.

[0110] Using P nThe voiceprint image is constructed by the formula (i,j)=B((n-1)M×N / 2+M(i-1)+j), which maps the normalized key intrinsic mode function amplitude to the pixel grayscale value of the RGB image. This formula means that for each pixel in the image, its RGB color value will be determined by the normalized amplitude of the three key intrinsic mode functions at the corresponding position. In this way, the information of different frequencies and intensities in the acoustic signal is encoded in the color of the image, forming a voiceprint image.

[0111] In the extraction unit, the construction process of the gray-level co-occurrence matrix includes: selecting a reference point (x, y) and a co-occurrence point (x+a, y+b) on the voiceprint image, so that the grayscale values ​​of the reference point and the co-occurrence point form a grayscale value combination (g1, g2); let the grayscale value level of the voiceprint image be k, transform the coordinates of the reference point and the co-occurrence point to obtain various grayscale value combinations (g1, g2), and the combination of the grayscale value combinations (g1, g2) has k 2 traverse all the reference points in the voiceprint image, count the occurrence frequency of each gray value combination (g1, g2), and obtain the frequency statistics result; organize the frequency statistics result into a square matrix, normalize the square matrix and convert it into a probability matrix P (g1, g2), the probability matrix is ​​the initial gray level co-occurrence matrix, wherein the rows and columns of the initial gray level co-occurrence matrix correspond to the gray values ​​g1 and g2 in the voiceprint image respectively; adopt the formula The above initial gray level co-occurrence matrix is ​​normalized, where p'(g1, g2, a, b) is the gray level co-occurrence matrix after normalization, a and b are the distance difference values ​​of the gray level values ​​g1 and g2 in the voiceprint image, N' is the number of gray levels, and R is the normalization factor.

[0112] Specifically, a grayscale co-occurrence matrix is ​​constructed to quantify the distribution and spatial relationship of gray levels in an image, thereby extracting statistics that describe the texture features of the image. On the voiceprint image, first select a reference point (x, y), and then select a co-occurrence point (x+a, y+b) that is a certain distance away from the reference point in a specific direction. Here, a and b define the spatial relationship between the two points, which can be horizontal, vertical, or diagonal offsets. The grayscale values ​​of the reference point and the co-occurrence point constitute a grayscale value combination (g1, g2), where g1 is the grayscale value of the reference point, and g2 is the grayscale value of the co-occurrence point. Since the grayscale value of the voiceprint image has k levels, there are a total of k possible grayscale value combinations. 2 Traverse the entire voiceprint image, for each pair of reference points and co-occurrence points, record the number of times their gray value combination (g1, g2) appears. This process needs to be performed for every possible reference point in the image, and finally a result of counting the frequency of occurrence of all gray value combinations is obtained.

[0113] The above frequency statistics are organized into a square matrix, where the rows of the matrix correspond to the grayscale value g1, the columns correspond to the grayscale value g2, and each element in the matrix represents the frequency of occurrence of the combination (g1, g2). Then, this square matrix is ​​normalized, that is, divided by the total number of pixels in the image to convert it into a probability matrix P(g1, g2). The normalized matrix elements represent the probability of the grayscale value combination (g1, g2) appearing in the image. Use the formula The probability matrix P(g1,g2) is further normalized to ensure that the sum of the matrix elements is equal to 1. Normalization is done by dividing each matrix element by a sum or factor (R) to ensure that the sum of the matrix elements statistically represents a probability distribution.

[0114] After completing the above steps, the normalized gray-level co-occurrence matrix p'(g1, g2, a, b) is a statistical matrix that quantitatively describes the relative frequency and spatial relationship of gray values ​​in the image. This matrix can be used to calculate a variety of texture features, such as contrast, energy, and entropy, to further reveal the intrinsic structure and characteristics of the voiceprint image, and provide valuable feature input for electrical equipment leakage detection based on voiceprint signals. In practical applications, since the construction of the gray-level co-occurrence matrix depends on the selection of reference points and co-occurrence points, different selection strategies and distance differential values ​​a and b will lead to different matrices. Therefore, multiple gray-level co-occurrence matrices are usually constructed in different directions and offsets, and then the average or maximum value features are extracted from these matrices to fully reflect the texture characteristics of the image.

[0115] The extraction unit comprises:

[0116] Extraction module for using formula The above contrast feature quantity is extracted from the above gray level co-occurrence matrix using the formula The above energy characteristics are extracted from the above gray-level co-occurrence matrix using the formula The above entropy feature is extracted from the above gray-level co-occurrence matrix, where i and j traverse the rows and columns of the above gray-level co-occurrence matrix, respectively representing a gray-level value in the above gray-level co-occurrence matrix, and p(i,j) is an element in the above gray-level co-occurrence matrix, indicating the probability of a pixel pair with gray-level values ​​​​i and j appearing at a given distance difference value (a, b).

[0117] Specifically, the texture features of the voiceprint image are extracted through the gray-level co-occurrence matrix, including contrast features, energy features and entropy features, which are used to quantify the texture properties of the image and provide feature input for subsequent image recognition and classification. The contrast feature, also known as contrast, directly reflects the contrast of the brightness of the image pixel values. It is proportional to the depth and clarity of the texture grooves. The larger its value, the deeper the grooves and the clearer the effect. The energy feature is an indicator of the stability of the grayscale change of the texture image, reflecting the fineness of the texture and the grayscale distribution. The energy value is equal to the square sum of the element values ​​in the matrix. The larger the energy value, the more stable the regular change of the texture. The entropy feature is an indicator of the amount of image information, reflecting the complexity of the texture in the image, and its value is proportional to the complexity of the image. When the pixel values ​​in the gray-level co-occurrence matrix show great randomness or all their values ​​are the same, the entropy value is the largest and the image is more complex.

[0118] By calculating the contrast feature, energy feature and entropy feature of the grayscale co-occurrence matrix, the texture structure of the voiceprint image can be quantified, providing key feature input for the image-based leakage detection method. These feature quantities can not only reveal the distribution and contrast of grayscale values ​​in the image, but also reflect the complexity and information content of the image, which is important for distinguishing normal images from faulty images. In the embodiment of the present application, these feature quantities are used to train and optimize the CNN-SVM model to improve the accuracy and robustness of leakage detection.

[0119] In the above-mentioned classification and recognition unit, the construction process of the above-mentioned CNN-SVM model includes: constructing a CNN model, the above-mentioned CNN model includes an input layer, a convolution layer, an activation layer, a pooling layer and a fully connected layer, wherein the above-mentioned input layer is used to take the above-mentioned voiceprint image as input, the above-mentioned convolution layer is used to automatically learn the above-mentioned texture features of the above-mentioned voiceprint image through a convolution kernel, the above-mentioned activation layer is used to increase the nonlinear expression ability of the network, the above-mentioned pooling layer is used to reduce the size of the above-mentioned voiceprint image, extract the high-level features of the above-mentioned voiceprint image, and at the same time reduce the calculation complexity and prevent overfitting, the above-mentioned fully connected layer uses a ReLU activation function to convert the two-dimensional feature map output by the above-mentioned convolution layer into a one-dimensional vector to obtain a feature vector; the above-mentioned feature vector is input into the SVM model for training to obtain the above-mentioned CNN-SVM model.

[0120] Specifically, the process of building a CNN-SVM model is to combine two machine learning models, Convolutional Neural Network (CNN) and Support Vector Machine (SVM), to efficiently extract features from voiceprint images and classify them. CNN is a feedforward neural network that contains convolution calculations and has a deep structure. The CNN network not only supports multi-layer neural networks, but also can learn feature parameters and classifiers at the same time. The structure of the CNN network is similar to a multi-layer perceptron, with a feedforward structure and the characteristics of weight sharing and local perception in image analysis. Its basic network structure includes 5 parts: input layer, convolution layer, pooling layer, fully connected layer and output layer.

[0121] The convolution layer is the core layer of the convolutional neural network. It processes the input image data through convolution operations to extract local features of the image. The essence of the convolution kernel is a weight matrix. In addition to the weight matrix itself, the number and step size of the convolution kernel are also important parameters. The corresponding number of convolution kernels are selected according to the different dimensions of the extracted feature matrix. The input size is proportionally reduced by the step size. That is, the convolution layer can change the size of the image and the number of channels of the image, as shown in the formula As shown, w1 is the width of the input image, h1 is the height, c1 and m represent the number of channels, k is the convolution kernel size, s is the operation step size, and p is the zero padding size, then w2, h2, and c2 are the width, height, and number of channels of the output image.

[0122] The pooling layer downsamples the image input by the previous layer, reduces the size of the matrix, and thus reduces the number of nodes in the fully connected layer, thereby reducing network parameters and avoiding overfitting. Commonly used pooling methods include average pooling and maximum pooling. The pooling layer only changes the size of the image, does not affect the depth of the three-dimensional matrix, and maintains the invariance of local features. Its mathematical model is as follows: As shown, w1 is the width of the input image, h1 is the height, c1 represents the number of channels, k is the convolution kernel size, s is the operation step size, p is the zero padding size, then w2, h2, c2 are the width, height and number of channels of the output image.

[0123] The fully connected layer is a traditional feedforward neural network structure in which each neuron is fully connected to all neurons in the previous layer. Its main function is to summarize the information input from the previous layer and act as a "classifier" in the neural network to classify the features extracted from the feature map of the previous layer. The fully connected layer usually uses the ReLU activation function and is usually located in the last few layers of the network to convert the two-dimensional feature map output by the convolutional layer into a one-dimensional vector to achieve an end-to-end learning process.

[0124] The output layer is the last layer of the neural network. Its function is to pass the output of the previous layer to the output function, usually using the logistic regression function (i.e. Softmax regression). The function of the Softmax function is to convert the output of the fully connected layer into a value between 0 and 1 and output the result in the form of probability. It is often used in classification tasks in convolutional neural networks. The Softmax function is defined as follows: As shown, where pi is the probability of the category output, the numerator is the exponential transformation of the output corresponding to the category, and the denominator is the sum of the exponential transformations of the output of the corresponding category, and n represents the category type.

[0125] VGG16 is a relatively mature convolutional neural network in the current CNN network. The VGG16 architecture is used as the CNN part of the network model, as shown in the formula input→[(conv×2→pool)×2]→[(conv×3→pool)×3]→FC×3. VGG16 uses the softmax function as the final output of the VGG16 architecture, but since the softmax classifier only performs probability conversion on the input of the previous layer, it does not have the ability to learn independently and cannot effectively improve the generalization ability of the image classification recognition model. Therefore, SVM is used instead of the softmax function to recalculate the values ​​in the final output layer.

[0126] The output feature vector of the CNN model is input into the SVM model for training. SVM is a supervised learning model used for classification and regression tasks. Its core is to find a hyperplane that can maximize the class interval to distinguish samples of different categories. In the application embodiment, SVM is used to learn the decision boundary between leakage and normal state based on the feature vector extracted by CNN. The training process of the SVM model involves finding the best classification hyperplane so that the model can accurately distinguish between voiceprint images with leakage and without leakage. Through training, the SVM model can learn the key features that describe leakage faults, which will be used for subsequent leakage detection and fault source location. After the SVM model training is completed, it can predict whether there is leakage based on the feature vector of the new voiceprint image.

[0127] In summary, the construction process of the CNN-SVM model is to automatically learn and extract the texture features of the voiceprint image through CNN, and then map these feature vectors to the decision space through SVM to achieve the classification of the leakage status of electrical equipment. This method that combines deep learning and traditional machine learning can not only automatically extract key features in the image, but also use the strong classification ability of SVM to perform accurate classification tasks, thereby improving the accuracy and efficiency of leakage detection.

[0128] In the above calculation unit, the location of the fault source is determined by using the above delay difference, including: using formula d i =v×(ti -t0) calculate the distance d between the above fault source and the i-th sensor i , where t0 is the time when the above leakage fault occurs, t i is the time when the i-th sensor receives the above sound signal, v is the speed of sound; solve the objective function Calculate the coordinates of the fault source in space, and use the coordinates of the fault source in space as the position of the fault source, where P is the position of the fault source, P i is the position of the i-th sensor, and D is the number of the acoustic signal sensors.

[0129] Specifically, the location of the sound source is determined based on the time difference of the sound signal arriving at different sensors, and the distance is estimated using the speed of sound and the time difference. Then, the time difference data of multiple sensors is used to solve the location of the sound source using triangulation or overdetermined equations. First, the relative distance between the fault source and each sensor is estimated by detecting the time difference of the sound signal arriving at each sensor. This is the basis for positioning using the time difference. Next, the objective function is used To determine the exact coordinates of the fault source in space, the objective function aims to minimize the difference between the actual distance and the estimated distance. By solving this optimization problem, an accurate position estimate of the fault source can be obtained. Generally, gradient descent, least squares method or more complex nonlinear optimization algorithms can be used to solve such problems.

[0130] The above-mentioned electrical equipment leakage identification device based on voiceprint signal includes a processor and a memory. The above-mentioned processing unit, generation unit, extraction unit, classification and identification unit, calculation unit, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize the corresponding functions. The above-mentioned modules are all located in the same processor; or, the above-mentioned modules are located in different processors in any combination.

[0131] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0132] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored program. When the program is running, the device where the computer-readable storage medium is located is controlled to execute the electrical equipment leakage identification method based on voiceprint signals.

[0133] An embodiment of the present invention provides an electronic device, the device comprising a processor, a memory, and a program stored in the memory and executable on the processor, and when the processor executes the program, at least the following steps are implemented:

[0134] Step S201, using a variational mode decomposition algorithm to process the collected acoustic signal to obtain a plurality of intrinsic mode functions, and selecting three key intrinsic mode functions from the plurality of the intrinsic mode functions;

[0135] Step S202, using the three key intrinsic mode functions as grayscale values ​​of the red, green and blue color channels of the image, respectively, to generate a voiceprint image of the acoustic signal;

[0136] Step S203, extracting texture features of the voiceprint image through a gray level co-occurrence matrix, wherein the texture features include contrast feature quantity, energy feature quantity and entropy feature quantity;

[0137] Step S204, using the CNN-SVM model to perform deep feature extraction and classification recognition on the above texture features to obtain leakage detection results;

[0138] Step S205 , when the leakage detection result indicates that a leakage fault occurs, the time delay difference of the acoustic signal propagating to the plurality of acoustic signal sensors is calculated, and the location of the fault source is determined using the time delay difference.

[0139] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program for initializing at least the following method steps:

[0140] Step S201, using a variational mode decomposition algorithm to process the collected acoustic signal to obtain a plurality of intrinsic mode functions, and selecting three key intrinsic mode functions from the plurality of the intrinsic mode functions;

[0141] Step S202, using the three key intrinsic mode functions as grayscale values ​​of the red, green and blue color channels of the image, respectively, to generate a voiceprint image of the acoustic signal;

[0142] Step S203, extracting texture features of the voiceprint image through a gray level co-occurrence matrix, wherein the texture features include contrast feature quantity, energy feature quantity and entropy feature quantity;

[0143] Step S204, using the CNN-SVM model to perform deep feature extraction and classification recognition on the above texture features to obtain leakage detection results;

[0144] Step S205 , when the leakage detection result indicates that a leakage fault occurs, the time delay difference of the acoustic signal propagating to the plurality of acoustic signal sensors is calculated, and the location of the fault source is determined using the time delay difference.

[0145] 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, 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.

[0146] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0147] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0148] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0149] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0150] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0151] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0152] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0153] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0154] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for identifying leakage of electrical equipment based on voiceprint signals, characterized in that: The electrical equipment is AC / DC hybrid power supply and distribution network electrical equipment, including: Processing the collected acoustic signal using a variational mode decomposition algorithm to obtain a plurality of intrinsic mode functions, and selecting three key intrinsic mode functions from the plurality of intrinsic mode functions; The three key intrinsic mode functions are used as grayscale values ​​of the red, green and blue color channels of the image respectively to generate a voiceprint image of the acoustic signal; Extracting texture features of the voiceprint image through a gray level co-occurrence matrix, wherein the texture features include contrast feature quantity, energy feature quantity and entropy feature quantity; The CNN-SVM model is used to perform deep feature extraction and classification recognition on the texture features to obtain leakage detection results; When the leakage detection result indicates that a leakage fault occurs, the time delay difference of the acoustic signal propagating to the plurality of acoustic signal sensors is calculated, and the location of the fault source is determined using the time delay difference.

2. The method according to claim 1, characterized in that: The collected acoustic signal is processed using a variational mode decomposition algorithm to obtain a plurality of intrinsic mode functions, and three key intrinsic mode functions are selected from the plurality of intrinsic mode functions, including: Apply formula 1 The original acoustic signal is decomposed into a set of modal components, where K is the number of modal components, {u k } is a set of modal components, {ω k } is the center frequency corresponding to the modal component, u k (t) is the kth modal component, ω k is the center frequency of the kth modal component, represents the derivative operation of the signal, j represents the imaginary unit, t is the time variable, * denotes the convolution operation, δ(t) is the Dirac function; For the formula 1, the Lagrangian multiplier λ and the penalty term α are introduced to construct the Lagrangian function Iterative optimization algorithm is used to solve u k ,ω k and λ(t), until the change of the modal component is less than a preset threshold, wherein λ(t) is a Lagrange multiplier function and f(t) is the original acoustic signal; The characteristics of all the modal components are analyzed, and three of the modal components containing key signal information are selected as the three key eigenmode functions.

3. The method according to claim 1, characterized in that The three key intrinsic mode functions are used as grayscale values ​​of the red, green and blue color channels of the image respectively to generate a voiceprint image of the acoustic signal, including: Normalizing each of the key intrinsic mode functions; Create a framework of an RGB image, wherein the RGB image consists of three color channels: red, green, and blue, and each color channel represents information of a different color of the RGB image; Using formula P n The voiceprint image is constructed by using the formula (i, j)=B((n-1)M×N / 2+M(i-1)+j), wherein P(.) represents the pixel grayscale value at the corresponding position of the RGB image, B(.) represents the normalized amplitude of the key intrinsic mode function, the size of the RGB image is M×N, i=1:M, j=1:N, and the normalized amplitudes of the three key intrinsic mode functions are respectively used as the grayscale values ​​of the red (R), green (G), and blue (B) color channels of the RGB image.

4. The method according to claim 1, characterized in that The construction process of the gray level co-occurrence matrix includes: Selecting a reference point (x, y) and a symbiotic point (x+a, y+b) on the voiceprint image, so that the grayscale values ​​of the reference point and the symbiotic point form a grayscale value combination (g1, g2); Let the grayscale value level of the voiceprint image be k, transform the coordinates of the reference point and the symbiotic point to obtain various grayscale value combinations (g1, g2), and the grayscale value combinations (g1, g2) have k combinations. 2 kind; Traverse all the reference points in the voiceprint image and count each gray value combination (g1, g2) to obtain the frequency statistics result; Arrange the frequency statistics results into a square matrix, normalize the square matrix and convert it into a probability matrix P(g1, g2), wherein the probability matrix is ​​an initial gray-level co-occurrence matrix, wherein the rows and columns of the initial gray-level co-occurrence matrix correspond to the gray-level values ​​g1 and g2 in the voiceprint image respectively; Using formula The initial gray level co-occurrence matrix is ​​normalized, wherein p'(g1, g2, a, b) is the gray level co-occurrence matrix after normalization, a and b are the distance difference values ​​of the gray values ​​g1 and g2 in the voiceprint image, N' is the number of gray levels, and R is the normalization factor.

5. The method according to claim 1, characterized in that Extracting the texture features of the voiceprint image through a gray level co-occurrence matrix includes: Using formula The contrast feature quantity is extracted from the gray level co-occurrence matrix using the formula The energy feature quantity is extracted from the gray level co-occurrence matrix using the formula Extracting the entropy feature from the gray-level co-occurrence matrix, Among them, i and j traverse the rows and columns of the gray-level co-occurrence matrix, respectively representing a gray value in the gray-level co-occurrence matrix, and p(i,j) is an element in the gray-level co-occurrence matrix, indicating the probability of a pixel pair with gray values ​​i and j appearing under a given distance difference value (a, b).

6. The method according to claim 1, characterized in that The construction process of the CNN-SVM model includes: Constructing a CNN model, the CNN model includes an input layer, a convolution layer, an activation layer, a pooling layer and a fully connected layer, wherein the input layer is used to take the voiceprint image as input, the convolution layer is used to automatically learn the texture features of the voiceprint image through a convolution kernel, the activation layer is used to increase the nonlinear expression ability of the network, the pooling layer is used to reduce the size of the voiceprint image, extract the high-level features of the voiceprint image, and reduce the computational complexity and prevent overfitting, and the fully connected layer uses a ReLU activation function to convert the two-dimensional feature map output by the convolution layer into a one-dimensional vector to obtain a feature vector; The feature vector is input into the SVM model for training to obtain the CNN-SVM model.

7. The method according to claim 1, characterized in that Determining the location of the fault source by using the delay difference includes: Using formula d i =v×(t i -t0) calculate the distance d between the fault source and the i-th sensor i , where t0 is the time when the leakage fault occurs, t i is the time when the i-th sensor receives the acoustic signal, and v is the speed of sound; Solving the objective function Calculate the coordinates of the fault source in space, and use the coordinates of the fault source in space as the position of the fault source, where P is the position of the fault source, P i is the position of the i-th sensor, and D is the number of the acoustic signal sensors.

8. An electrical equipment leakage identification device based on voiceprint signals, characterized in that: include: A processing unit, configured to process the collected acoustic signal using a variational mode decomposition algorithm to obtain a plurality of intrinsic mode functions, and select three key intrinsic mode functions from the plurality of intrinsic mode functions; A generating unit, configured to use the three key intrinsic mode functions as grayscale values ​​of red, green and blue color channels of an image respectively, to generate a voiceprint image of the acoustic signal; An extraction unit, configured to extract texture features of the voiceprint image through a gray level co-occurrence matrix, wherein the texture features include a contrast feature quantity, an energy feature quantity, and an entropy feature quantity; A classification and recognition unit, used for performing deep feature extraction and classification and recognition on the texture features using a CNN-SVM model to obtain a leakage detection result; The calculation unit is used to calculate the time delay difference of the acoustic signal propagating to multiple acoustic signal sensors when the leakage detection result indicates that a leakage fault occurs, and determine the location of the fault source using the time delay difference.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method for identifying leakage of electrical equipment based on voiceprint signals according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a method for executing the electrical equipment leakage identification method based on voiceprint signals as described in any one of claims 1 to 7.

Images